I recently toured Ginkgo Bioworks’ HQ in Boston’s Seaport District and sat down with CEO Jason Kelly for a deep-dive interview. Ginkgo is placing a big bet on the future of life science being autonomous. Here’s why I think it’s an area worth paying attention to.

Distilling Drug Discovery

Taking a step back, what does it actually take to develop a cure for a disease? In my research, it boils down to three moves: Identify the disease you wish to treat. Develop an informed hypothesis about the mechanisms by which it unfolds in the human body – an extraordinarily complex, not-yet-fully-understood system of over 30 trillion cells. Then develop and manufacture a therapy that influences those mechanisms to alleviate symptoms, with minimal side effects.

In the US, this process famously costs – on average – over $2.5 billion and takes a decade or longer. So, how do we make it cheaper and faster?

One approach: enable the entire system to run more efficiently. Tedious note-taking? Electronic lab notebooks let researchers go digital-first. Repetitive pipetting? Automated liquid handlers help lighten the load. High labor costs? Contract Research Organizations offer economies of scale and, when offshored, cheaper wages.

Globally, policy innovation is also an important lever for efficiency: for example, in the US, clinical trials require a high (and expensive) burden of proof to be greenlit by the FDA, standing in contrast to countries such as Australia and China, where an “implicit approval” system approves trials automatically unless a safety concern triggers intervention.

Better fundamental research also helps. More accurate models of disease and human biology mean more informed hypotheses, more shots on goal, and thus more effective use of every research dollar. Long-term initiatives like the Human Genome Project have spawned new categories of treatment, such as promising gene therapies.

And yet, cost and timelines have remained stubbornly high. The industry even has a name for it: Eroom’s Law – Moore’s Law spelled backwards – which holds that “the inflation-adjusted cost of developing a new drug roughly doubles every nine years.” Meanwhile, large pharma companies are increasingly turning to China to license sophisticated new drugs that were once the hallmark of American innovation.

The Case for Autonomous Life Science

What if we gave AI access to a wet lab?

At a basic level, an embodied AI system could surface new efficiencies in how science is done – optimizing expensive reagent use, running experiments in parallel, reducing equipment idle time. More ambitiously, automation could bring back contract research work that’s been offshored, recapturing lost margin and, perhaps more importantly, protecting sensitive IP from leakage.

Looking further ahead: with enough experience interfacing with real-world biology, could AI models develop better-informed hypotheses about how diseases unfold – and which therapies to test? Could these systems extend beyond the lab to advise on patient recruitment, trial design, and even full R&D strategy to further reduce the cost and time it takes to bring new drugs to market?

This vision sits at the heart of Ginkgo Bioworks, one of a number of companies leaning hard into autonomous science. Ginkgo has seen its share of ups and downs since going public in 2021, but the company has recently notched two meaningful wins. First, a $47 million contract to augment the capabilities of the DOE’s Pacific Northwest National Lab’s Microbial Molecular Phenotyping Capability (M2PC), focused in part on generating rich microbial datasets. Second, a recent partnership with OpenAI, where they collaborated on leveraging GPT‑5 to lower the cost of cell-free protein synthesis, claiming a 40% reduction in protein production cost.

What follows is my conversation with Jason on his vision for autonomous life science, the challenges of realizing it (both scientific and economic), and the geopolitical forces driving AI-for-science investment.

If you’re curious about where frontier AI meets the lab bench, check out the latest episode of the Discovery Engines Podcast.

Watch or Listen Now

Also available on:

Spotify:

Apple Podcasts:

Show Notes

Episode Links:

• OpenAI x Ginkgo cell-free protein synthesis collaboration: https://openai.com/index/gpt-5-lowers-protein-synthesis-cost/

• Genesis Mission to Accelerate AI for Scientific Discovery: https://www.whitehouse.gov/fact-sheets/2025/11/fact-sheet-president-donald-j-trump-unveils-the-genesis-missionto-accelerate-ai-for-scientific-discovery/

• PNNL’s Anaerobic Microbial Phenotyping Platform (AMP2): https://www.emsl.pnnl.gov/science/instruments-resources/anaerobic-microbial-phenotyping-platform-amp2

• Ginkgo’s win for PNNL’s Microbial Molecular Phenotyping Capability (M2PC): https://www.prnewswire.com/news-releases/ginkgo-bioworks-selected-by-pnnl-to-deliver-a-modular-highthroughput-phenotyping-platform-for-does-m2pc-302634071.html

• Society for Lab Automation and Screening: https://www.slas.org/

• Globaldata Report on Large pharma drug licensing from China: https://www.globaldata.com/media/business-fundamentals/large-pharma-drug-licensing-from-china-reaches-record-high-at-28-in-2024-reveals-globaldata/

• Ginkgo Bioworks: https://www.ginkgo.bio/

• Jason Kelly: https://www.linkedin.com/in/jrkelly2/

Episode Chapters:

• 01:54 - Jurassic Park and the Value of Storytelling in Science

• 04:50 - What Is the Department of Energy's AMP2 at PNNL?

• 07:17 - How Autonomous Labs Can Be Used to Run Experiments

• 11:49 - Principles of the Autonomous Lab (Waymo Analogy)

• 17:12 - Project Genesis, Competing with China in Scientific Discovery

• 23:42 - Reindustrializing America via Automation

• 25:52 - Challenges to Achieving the Autonomous Lab

• 37:03 - OpenAI Designing and Running Bio Experiments

• 47:17 - Advice to Frontier Labs Designing AI Intelligence for Science

• 49:24 - Ginkgo Autonomous Lab Tour

• 1:00:38 - OpenAI's Interface With Ginkgo

• 1:02:51 - Lab Scheduler Platform

• 1:16:19 - Advice for Deep Tech Startups

• 1:22:22 - Advice for Scientists, Founders Thinking About Where to Focus in Biotech

• 1:25:18 - How Research Labs Could Better Collaborate and Communicate Results

• 1:26:35 - The Pharma Industry Takes Exploratory Risks

• 1:29:23 - Advantages and Challenges of Going Public

• 1:32:14 - Public Perception of Ginkgo

• 1:33:57 - How Ginkgo's Business Model Has Changed

Imagine you’re eight years old, born into a wealthy family that owns a sprawling, highly lucrative enterprise. Suddenly you’re orphaned, and responsibility for this vast inheritance falls to you.

As a first step you - smartly - search for an adult to manage the estate until you’re older. You interview candidate after candidate but you keep running into the same question: how do you know who to trust? You want someone who genuinely prioritizes your long-term interests. Several appear sincere, yet it’s impossible to tell who’s truly aligned with you and who’s simply skilled at acting aligned. Or worse: someone who will obey your requests blindly even when those requests drive the company off a cliff.

It’s this analogy at the heart of “Why AI Alignment Could Be Hard with Modern Deep Learning”, a thought-provoking essay and that my learning group read as part of an AI Safety Fundamentals course I recently completed with MIT AI Alignment (MAIA).

In the spirited discussions that followed, we covered a number of strategies, from developing ways to “control” superior intelligences, to inspecting their thought process through techniques such as mechanistic interpretability and probing. My personal take-away: this stuff is really hard. There’s something inherently slippery about trying to control a mind that could be vastly more intelligent than our own, especially if the time-scales for control extend beyond our lifespans.

It’s these questions, and the larger landscape of AI safety, that set the theme for my recent sit-down with Riya Tyagi and Gatlen Culp, MIT undergrads and MAIA board members. In our conversation, we covered:

• Emerging AI safety techniques

• Political and economic drivers of existential risk

• Future-state scenarios (like AI-powered autonomous slaughterbot drones)

• Generational differences in how people view AI

• Concrete ways to get involved in and support the AI safety community

Fun fact: Riya is a Mechanistic Interpretability Researcher at the lab of Max Tegmark, well-known AI researcher, safety advocate, and author of Life 3.0, a book that first opened my eyes to the possibilities - and dangers - of superintelligent AI systems.

If you’re curious about the cutting edge of AI safety research, from the perspective of the next generation of AI researchers, this conversation is for you.

Watch or Listen Now

Also available on:

Spotify: https://open.spotify.com/show/1wfOlAAqXXaIrXRwO5T71R
Apple Podcasts: https://apple.co/40sQCu9

Show Notes

Episode Links:

• MIT AI Alignment: https://aialignment.mit.edu/

• Cambridge Boston Alignment Initiative (CBAI): https://www.cbai.ai/

• Riya’s LinkedIn: https://www.linkedin.com/in/riyatyagi-ai/

• Gatlen’s LinkedIn: https://www.linkedin.com/in/gatlen-culp/

• Gatlen’s Projects: https://www.mit.edu/~gculp/

• Slaughterbots x Future of Life Institute: https://futureoflife.org/video/slaughterbots/

• Track II Diplomacy: https://en.wikipedia.org/wiki/Track_II_diplomacy

• Tegmark AI Safety Group: https://tegmark.org/

• “Nexus” by Yuval Noah Harari: https://www.ynharari.com/book/nexus/

• RAND Institute: https://www.rand.org/topics/artificial-intelligence.html

• BlueDot Impact: https://bluedot.org/

• “If Anyone Builds It, Everyone Dies” by Eliezer Yudkowsky and Nate Soares https://en.wikipedia.org/wiki/If_Anyone_Builds_It,_Everyone_Dies

Episode Chapters:

• 00:00 - Preview & Intro

• 01:48 - What Is MIT AI Alignment?

• 02:47 - Why AI Safety?

• 09:26 - Trends in AI Safety Interest

• 12:46 - AI Safety Techniques: MechInterp & Beyond17:15 - Model Situational Awareness

• 20:58 - Hybrid, Mixture of Experts Models

• 24:40 - Decomposing a Model & Parallels with Human Brains

• 29:00 - Private Capital for Safety Research

• 32:23 - Frontier Lab Mentorship Programs

• 34:14 - Policy Perspectives & China Competition

• 36:53 - Ways In Which AI Might Threaten Us

• 39:31 - Track 2 Diplomacy & International Collaboration Examples

• 43:13 - Slaughterbots & Dangerous Capability Demos

• 46:54 - AI-Driven Unemployment

• 52:06 - Generational Attitudes Towards AI

• 57:44 - How to Get Involved - Non Technical

• 01:01:00 - Learning Resources: BlueDot Impact, etc

• 01:01:46 - Importance of Communicators & Artists

• 01:03:50 - How to Support MAIA

Over 100 years ago, Albert Einstein developed relativity, which revealed profound truths about the interconnectedness of space and time, and the vast amount of energy released when mass is converted into energy. This insight laid the foundation for a new generation of technologies, including the nuclear power plants that today provide 10% of the earth’s electricity.

Around this time, scientists such as Max Planck, Niels Bohr, Werner Heisenberg, Erwin Schrödinger, Paul Dirac and others pushed physics into new territory, revealing the “quantum” nature of reality at its smallest scales. Discoveries that are still being built upon today, to design everything from microprocessors, to emerging quantum computers and sensors that promise step-changes in computational efficiency, medical imaging, and navigation.

And yet, by some measures, the pace of discovery in fundamental physics has slowed. In 2012 humanity’s largest particle collider - a 27 km loop based in Geneva, Switzerland - confirmed the existence of the Higgs Boson – further shoring up the Standard Model of Physics – but leaving questions like quantum gravity unanswered. String theorists have proposed potential explanations, but face a landscape of plausible formulations that remain difficult to narrow down.

Enter Jesse Thaler. Director of the National Science Foundation’s Institute for Artificial Intelligence and Fundamental Interactions (IAIFI), and theoretical particle physicist at MIT, I recently had a chance to sit down with him and get a peek into the work he and his colleagues are doing to accelerate physics discovery with AI. From maximizing the discovery potential of data-generated from the Large Hadron Collider, to building companion tools that help physicists amplify their thinking.

In my hour-long conversation with Jesse we covered a range of topics, including:

• The interplay between physics research and advancements in AI (diffusion models, efficient scaling, and beyond)

• The current and future state of particle colliders like the LHC

• The essential role of curiosity-driven research

• Jesse’s “hot takes” on quantum mechanics, Many-Worlds, and the nature of reality

If you’re curious about the cutting edge of physics, or how physics is helping address major challenges in AI – like making large language models scale more efficiently – this conversation is for you.

Watch or Listen Now

Also available on:

Spotify: https://open.spotify.com/show/1wfOlAAqXXaIrXRwO5T71R
Apple Podcasts: https://apple.co/40sQCu9

Show Notes

Episode Links:

• Jesse's Personal Website: https://jthaler.net/

• National Science Foundation's IAIFI: https://iaifi.org/

• MIT Department of Physics https://physics.mit.edu/

• Jesse's MIT Faculty Page: https://physics.mit.edu/faculty/jesse-thaler/

• Quantum Field Theory: https://en.wikipedia.org/wiki/Quantum_field_theory

• CERN Large Hadron Collider: https://home.cern/science/accelerators/large-hadron-collider

Episode Chapters:

• 00:00 - Preview & Introduction

• 04:00 - Jesse's Journey Into Physics

• 05:49 - Curiosity For Understanding The Nature Of Reality

• 07:27 - What Is A Theoretical Particle Physicist?

• 09:34 - Quantum Field Theory

• 11:05 - Particle-Wave Duality

• 12:53 - What is the Large Hadron Collider?

• 15:04 - Data Generated From Particle Collisions

• 17:52 - How LHC Experiments Are Scheduled

• 19:24 - The Future of the LHC and Particle Colliders

• 26:29 - National Science Foundation's Institute for AI and Fundamental Interactions (IAIFI)

• 30:52 - How AI Is Accelerating Physics Discovery

• 40:42 - How Physics is Accelerating AI Discovery (Diffusion models and beyond)

• 48:27 - Physics Popcorn! – Einstein on Time

• 49:32 - Donald Hoffman's Interface Theory of Perception

• 50:48 - Many-Worlds Theory of Quantum Mechanics

• 53:12 - DeepMind's Demis Hassabis on Superintelligence for Science

• 54:03 - Human-Style Intelligence Alone Is Not Enough

• 55:17 - Advocate For Curiosity-Driven Research, Hire IAIFI Talent, Engage With AI

I recently landed in Cambridge, Massachusetts to begin my next chapter of AI for Science, with a deeper focus on accelerating physics research. In this post, I lay out the moral, philosophical, and economic case for accelerating physics discovery (warning: it gets trippy 🤡). Let’s go.

Consider, for a moment, the following scenarios:

Scenario 1: You’ve been playing a world-building video game like SimCity or Minecraft. You’ve invested considerable time and effort developing your now thriving city. As usual, you boot up the game to unwind after work or school. Today, however, the home screen catches your eye.

Rather than your single saved profile, you see seemingly infinite copies of it. Clicking into a few at random, you discover different versions of the world you created, each slightly modified, but all unmistakably designed by you.

Scenario 2: Imagine you’re playing the same game, but this time you don’t encounter the duplicate profile “bug.” The first time you play the game, after 20 hours of following your instincts, you build a world you’re really proud of. Called “New Block City,” it features skyscrapers with glass facades and rooftop gardens. And a happening entertainment district called “Pluto Plaza,” named after your childhood pup.

Wanting another go, you delete your world and start from scratch. You lose track of time as you lean in to building something that expresses your creativity. After another 20 hours, you zoom out to admire your progress and notice something peculiar: this world is eerily similar to the last. Upon closer inspection, your heart jumps: you’ve built an exact replica.

Brushing this off as coincidence, you delete your world again and rebuild from scratch. Yet each time, after 20 hours, you step back, and there it is: New Block City, featuring Pluto Plaza. Just like last time. And the time before that.

Scenario 3: Imagine you are the duplicated profile – and repeatedly rebuilding New Block City is your lived experience.

Into The Wilderness

One of humanity’s hallmarks is our ability to think beyond immediate needs and grapple with larger questions about ourselves and the nature of reality.

In the American education system, we often learn popular takes on these questions from a Western perspective. For example, René Descartes’ declaration, “I think therefore I am,” arguing that our capacity for thought is proof of existence, even if our senses deceive us. Or Plato’s Allegory of the Cave, which suggests that if all we’ve ever seen are shadows cast on a wall, we might mistake those illusions for reality itself.

Elsewhere in the world, we may be more exposed to the teachings of Zhuangzi, a Chinese philosopher, who after a deep slumber wondered, “Am I Zhuangzi who dreamed I was a butterfly, or a butterfly now dreaming I am Zhuangzi?” Or to Hinduism’s sacred texts, the Upanishads, which teach that consciousness plays a fundamental role in how we perceive reality.

A major breakthrough has been our ability to pair these philosophical questions with experimentally proven theories. Two relevant areas being the study of consciousness, through neuroscience, and the study of the world around us, through physics.

The advances we’ve made in these areas have been breathtaking. Rooted in a classical worldview, where time was considered absolute, our understanding lurched forward when Einstein revealed that time itself is relative and dependent on relative velocity and the forces exerted by gravitational fields. We took another huge step forward with the development of quantum mechanics, which revealed that physical systems are described by wave functions that encode probabilities, and which likely “collapse” upon observation and/or continuously branch into parallel worlds.

These discoveries gave birth to a new generation of practical technologies, from nuclear power plants that generate electricity, to accurate GPS clocks that help us navigate to our friend’s birthday party.

Yet they’ve also revealed how unintuitive reality is compared to our lived experience: a twin returning from space younger than the one who stayed on Earth? A single photon of light interfering with itself? And raised new scientific, and I would argue, spiritual questions.

If the wave-function of the universe does indeed evolve into many branching realities, this means that since you started reading this post, countless versions of you have split into different branches of the universe, each experiencing a distinct path of events. You are one of your saved gaming profiles. And each separate profile is another fully-formed, conscious version of you, skin, bone and spirit.

Or, per Einstein’s theory of relativity, there is no single universal “now,” only different slices of spacetime seen by different observers and dependent on their motion. Some scientists and philosophers interpret this to mean that past, present, and future must all equally exist in a “block universe.” The version of you who recently started reading this blog post exists as a fixed part of spacetime, just as the version reading this sentence exists now, each experiencing its own moment as the present, and ultimately rooted in a single, unchangeable story. Hello Pluto Plaza.

The Moral Opportunity

A few years ago, I was moved by an article about a restorative justice program in the US. In it, the parents of a child who was murdered by a young man chose to build a relationship with their child’s assailant, setting him on a path toward reform rather than seeking the death penalty. To me, it was a powerful example of the belief that we are born inherently good-hearted, shaped into decisions that help or harm others by social and physiological forces bigger than ourselves. It’s a belief mirrored in the teachings of many faiths and reflected in the US justice system, which begins with a presumption of innocence.

If each of our life choices is laser etched into the “DVD movie” of our existence, as implied by the block universe theory, what does that imply for our system of justice? If the future is deterministic, how might we design systems that account for an apparent lack of free will?

Or consider the interpretation of quantum physics in which infinite copies of each of us exist in separate realities. In the sum total of realities, each of us is undoubtedly both villain and victim, victor and vanquished. Does our empathy extend to those other selves? Do they hold a place in our thoughts and prayers? Or do we simply ignore them because they exist in different - albeit equally real - realities? How might we “update” our spiritual practices to reflect such deeper truths? (Shout-out to Emily Qureshi-Hurst, who recently penned the book: Salvation in the Block Universe).

The Economic Opportunity

Materials, Energy, and More

The technological opportunities arising from progress in physics are equally staggering. As we work to develop better energy storage systems or room-temperature superconductors, one of the largest barriers is the computational intractability of formulating the right chemistries.

There are roughly 90 naturally occurring elements on the periodic table. Choosing just six from a pool of 30 viable ones yields over half a million possible combinations. These calculations are intractable for even our most powerful classical computers, but far more manageable for a computer that leverages the properties of quantum physics to conduct massively parallel computations.

Conversely, losing our edge in these fields could make secure communications, banking, and other “password-protected” services vulnerable to brute-force hacking. And we risk missing out on the commercialization of transformative technologies that could arise from harnessing new physics: from more accurate sensors for medical imaging, to better controlled nuclear fusion for plentiful, cleaner energy.

Artificial Superintelligence

Despite all the progress LLMs and generative AI have made in recent years, they still struggle to extrapolate beyond their training data to new, unseen problems. It’s these types of insights that have driven some of our biggest breakthroughs as a species and, so far, our AI systems seem to fall short.

Designing systems that can make tangible progress in theoretical physics, whether through better analysis of existing data from high-energy physics, astronomy, and self-driving labs, or by leaping from existing theories to new ideas, might be the real “Turing test” for the superintelligent systems many labs are developing.

Progress also flows in the other direction. Many of today’s most impressive examples of generative AI, like image synthesis and protein prediction, are rooted in well-established physics methodologies, such as modeling the diffusion of gases in thermodynamics. What other physics techniques might “translate” to AI if further developed?

Building Community

Opportunities to simultaneously advance deeply philosophical and economically impactful questions are rare. It’s for this reason that I made the leap into AI for Science, with a special focus on physics.

The years I spent getting my MBA at MIT Sloan were in retrospect deeply impactful in planting the seeds for my journey. What I loved about my time at Sloan was the encouragement we received to get out of our building and build relationships with innovators and scientists across campus – to become partners in unlocking the promise of their ideas and making them real.

As I begin this next chapter, I’m excited to do it from the Cambridge/MIT ecosystem and connect and collaborate with the many excellent people and organizations within and around it, be it the MIT Center for Theoretical Physics, the Kavli Institute, NSF’s Institute for AI and Fundamental Interactions, or places like the Lincoln Lab that are putting theory into practice.

If you know folks at these organizations – or others in the Cambridge/Boston ecosystem who I should connect with – please drop me a line! And, as always, I would love to hear about your own projects and interests at the intersection of AI and physics.

Onward.

- Nabil

Enjoying this read? Subscribe for more and share it with friends eager to accelerate scientific discovery.

Subscribe now

I recently spent a week at the BIO International Convention in Boston, billed as the world’s largest and most comprehensive event for the biotechnology industry, with over 20,000 registrants from more than 70 countries. Through attending small panels, and roaming the cavernous exhibition floor, I learned a ton.

More on that (+ a video walkthrough) in “Deep Dive,” below. But first, some industry updates 👀

• DeepMind Introduces AlphaGenome: Mutations in the BRCA1 or BRCA2 genes are associated with a higher incidence of breast cancer. One of the holy grails of genomic research is understanding how the countless other mutations in our DNA influence biological processes and, ultimately, human health. To that end, DeepMind’s new AlphaGenome model takes DNA sequences as long as one million letters and predicts thousands of molecular properties characterizing its regulatory activity.

• Arc Institute Introduces The Virtual Cell Challenge: In parallel to DeepMind’s efforts, the Arc Institute is looking to better understand how our cells respond to genetic and chemical perturbations, key to unlocking human health. To do so, the Arc Institute has launched the Virtual Cell Challenge, awarding $175K in prizes to the top models that can predict the effects of perturbing 300 genes.

• MIT Researchers Enable Quantum Processors To Talk To Each Other Directly: The combinatorial explosion of developing novel materials, or simulating biological systems, overwhelms even our best “classical” supercomputers quickly. Quantum computers offer a way out, however they’re stubbornly sensitive to perturbations from the external environment. Point-to-point connectivity between quantum processors increases error rates. In support of a more direct (and less error prone) architecture, MIT researchers developed a new interconnect device that can support scalable, “all-to-all” communication, so that all superconducting quantum processors in a network can communicate directly with each other.

• AI-powered Engineering Platform PhysicsX Raises $135M Series B: It’s one thing to develop a novel material or technology, it’s another matter to scale up its production. Addressing this challenge head-on, UK-based startup PhysicsX recently raised a large round to develop AI software for advanced manufacturing. Here’s their materials value prop: "By refining component design across production lines — enhancing furnaces, mixing baths, heat exchangers, conveyor belts, and feeders — we drive efficiency and innovation. AI-powered process control maximizes production, minimizes emissions, and reduces maintenance."

During my week at the BIO International Convention, some of the most compelling conversations happened in the smaller, more intimate panels above the convention floor. A few I attended include:

• AI in Biologics Drug Discovery

• Computational Drug Design

• Supercharging Biomanufacturing Infrastructure

• How to Best Scale Cell & Gene Therapies

• Clinical Trials for Rare Diseases

• Bridging the Biodata Gap through “Digital Twins”

As I reflect on these – and the hallway conversations – a few themes stand out:

AI Beyond Drug Discovery

As AI generates ever-more molecules to test as potential drug candidates, it exacerbates bottlenecks downstream. Several industry leaders described the need to apply AI to the long tail of workflows that follow target or molecule identification, including in manufacturing. For example:

• Streamlining paper-based Batch Production Records (BPRs)

• Leveraging digital twins to optimize automation and yields

Innovation in a Regulated System

Biotech’s structural hurdles – regulatory complexity, long timelines, capital intensity – remain. But some panelists offered creative workarounds. One neat idea: Hospital systems developing and delivering their own cell and gene therapies in-house. Think “hospital as biotech startup,” vertically integrating development and care. It struck me as a concept right out of Doblin’s Ten Types of Innovation framework – one of my favorites.

Overbuilt Today, Platforms of Tomorrow?

The BIO convention floor was teeming with CDMOs, CROs, and other service providers. I spoke to one who described their rapid scale-up to revenue and profitability during the COVID era, now followed by a sharp drop in demand.

I wonder if there’s a parallel to the dot-com era, when over-investment in broadband infrastructure led some companies to shut down, but ultimately laid the foundation for a new wave of high-tech giants. Could what was overbuilt in the last cycle end up fueling the next?

Watch Now

For an inside peek at the convention, including interviews with those building organs-on-chips and automated liquid biopsies, check out my walk-through and take-aways here:

Featured Orgs & People

• BloodScan Biotech (Featuring Eric Lin)

• SCHOTT Pharma (Featuring Andrew Crawford, PhD)

• Central European BioForum (Featuring Magdalena Kulczycka, PhD)

• UK Department for Business and Trade (Featuring Raffaella Harvey)

• Choose New Jersey (Featuring Rachel Cohen, MBA)

• MEPSGEN (Featuring Ville Kujala)

• ReciBioPharm (Featuring Aaron Cowley)

Find this newsletter valuable? Subscribe for regular insights and share it with friends who are passionate about accelerating scientific discovery.

Subscribe now

Help us improve. Like a good neural network, this newsletter is only as effective as the input it receives. We’d love to hear your feedback – let us know how we can make it even better for you.

Share feedback

Catch you on the next one 🤙🙌 – Nabil

When we see ads from big biotech or pharma companies, often they showcase sleek labs filled with sophisticated robots precisely moving around small amounts of liquid.

These impressive lab machines come with extensive product lines that have accelerated science R&D. And, they're often expensive and tied to proprietary tech stacks. This creates barriers, especially for cash-strapped startups and bootstrapped scientists looking to innovate by mixing and matching different lab equipment, or experimenting with second-hand devices.

In this episode of Discovery Engines, I sit down with Rick Wierenga, creator of PyLabRobot, an open-source, hardware-agnostic Python SDK designed to tackle this challenge head on. What began as a spark of insight during a college biology class took an big leap forward when Rick got a surprising reply to a cold email he sent during a visit to MIT. Rick and a growing community of contributors are on a mission to remove barriers to lab automation, making it easier to “ship science.” For Rick’s journey, and insights into the growing movement behind PyLabRobot, join us for this conversation.

Watch or Listen Now

Also available on:

Spotify: https://spoti.fi/3BzMqiy
Apple Podcasts: https://apple.co/40sQCu9

Additional Show Notes

Episode Links:

• Rick's Twitter https://x.com/rickwierenga

• Rick's LinkedIn https://www.linkedin.com/in/rickwierenga/

• PyLabRobot GitHub https://github.com/PyLabRobot

• PyLabRobot Forum https://discuss.pylabrobot.org/

• PyLabRobot Twitter https://x.com/PyLabRobot

• Dana Gretton, PyHamilton https://github.com/dgretton

• PyTorch https://pytorch.org/

• Flow Cytometry https://en.wikipedia.org/wiki/Flow_cytometry

• Autonomous Science Opportunity Map https://bit.ly/autonomous-science-map

Episode Chapters:

• 00:00 - Preview & Introduction

• 03:33 - Rick's Serendipitous Foray Into Lab Automation

• 07:00 - Why AWS for Biology

• 09:17 - Understanding Liquid Handling Robots

• 10:49 - Generic Robotics Stack

• 13:20 - Robot Automation With PyLabRobot

• 15:07 - The Community Building PyLabRobot

• 16:54 - Why A Big Pharma Would Support PyLabRobot

• 18:36 - PyTorch: An Open Source Inspiration

• 20:32 - Sharing Technology In A Competitive Industry

• 23:11 - Supported Device Ecosystem

• 25:49 - Surprises in Device Integration

• 27:54 - Reverse Engineering 101

• 31:26 - How Useful Are Github CoPilot, Codex, LLMS for Reverse Engineering

• 34:34 - How Device Manufacturers, Automation Experts, Are Reacting to PyLabRobot

• 39:17 - Open Source Efforts and Device Safety

• 41:11 - How PyLabRobot Enables Sharing Protocols, Reproducibility

• 45:03 - Creating a 3D Scanner From a Liquid Handler

• 48:49 - Supporting Self-Driving Labs

• 54:49 - How PyLabRobot Will Evolve

• 55:40 - Automating Flow Cytometry

• 58:17 - How To Contribute and Connect

The biotech, chemicals, and materials industries offer so much potential to improve lives – whether through life-saving therapies, or breakthrough materials for a more sustainable planet.

But for many, entering these fields feels intimidating. Students face intense coursework and long hours in the lab. For professionals, pivoting into these domains can feel even harder.

That’s why I found Alishba Imran’s story so inspiring: an undergrad at University of California, Berkeley, she carved her own path through study groups, shadowing researchers, and cold outreach.

A deep learning researcher with Berkeley Artificial Intelligence Research (BAIR), Alishba now splits her time between completing her undergrad studies, and advancing AI for biology as a Research Fellow at the Arc Institute. She's also the author of a new book: "AI for Robotics: Toward Embodied and General Intelligence in the Physical World."

In my hour-long chat with Alishba, we explored:

• How she bootstrapped her learning across biology and chemistry

• What it takes to launch a deep-tech startup like Voltx

• Alishba's work advancing cell biology with self-supervised learning

• The promise – and hype – of robotics research today

• And how nonprofits like the Arc Institute are reshaping the landscape for researchers.

Watch or Listen Now

If you’re curious about the future of AI, biology, and robotics, this episode’s for you.

Also available on:

Spotify: https://spoti.fi/3BzMqiy
Apple Podcasts: https://apple.co/40sQCu9

Additional Show Notes

Episode Links:

• Alishba's LinkedIn https://www.linkedin.com/in/alishba-imran-/

• Alishba's Twitter https://x.com/alishbaimran_/

• "AI for Robotics" book https://link.springer.com/book/10.1007/979-8-8688-0989-7

• Arc Institute https://arcinstitute.org/

• UC Berkeley BAIR AI Lab https://bair.berkeley.edu/

• Chan Zuckerberg Biohub https://www.czbiohub.org/

• DynaCLR Paper https://arxiv.org/html/2410.11281v1

• Biopunk Lab https://biopunklab.com/

Episode Chapters:

• 00:00 - Preview & Introduction

• 05:02 – Why AI for Science?

• 10:34 – Bootstrapping Your Own Learning

• 16:02 – Why Battery Testing Matters: Launching Voltx

• 20:19 – Batteries, Raw Earth Minerals: Incremental vs Transformative Chemistries

• 23:54 – The Business Side of Deep Tech Entrepreneurship

• 29:14 – The State of Battery Tech Today

• 30:50 – Working with Chan Zuckerberg Biohub

• 33:37 – DynaCLR For Contrastive Learning of Cell State Dynamics

• 36:21 – Why Self-Supervised + Contrastive Learning

• 41:06 – Generalizing to Different Cells Contexts With Embeddings

• 42:06 – Microscopy or ML Advancements for Unlocking Progress

• 43:54 – Managing Terabytes of Data

• 46:51 – Arc Institute: Collecting Largest Single Cell Dataset

• 48:40 – Role of Tech-Founder Funded Institute To Advance Foundational Work

• 50:41 – Book Launch! Merging Classical Robotics Methods With Cutting-Edge Deep Learning

• 53:42 – Robotics Progress: Hype vs Reality

• 57:09 – Robotics for Science

• 58:54 – Future Trends: Automation, Foundational Single Cell Models, Protein and Genome Language Models, Transcriptomics

• 01:00:41 – Pivoting into AI for Science Early, Mid, or Late Career

• 01:02:49 – Exchanging Ideas; DynaCLR Paper; AI for Robotics Book

"Anytime you ask a question, that should be contextualized in all the data that you've generated."

I love this vision on the future of scientific discovery shared by Nicholas Larus-Stone, CEO of Sphinx Bio and founder of the Bits in Bio community.

In our wide-ranging sit-down conversation, we explored:

• What an engineering mindset uniquely brings to biotech

• The real-world data workflows of scientists today (shout-out to thumb drives!)

• The transformative potential of AI co-scientists

• Managing AI hallucinations and building trust

• The story of Bits in Bio: the community that’s helping shape the future of Software x Science

Watch or Listen Now

If you're excited about the intersection of AI, agents, data, and science, and want a glimpse into the future of discovery, join us for this conversation.

Also available on:

Spotify: https://spoti.fi/42Uvv5J
Apple Podcasts: https://apple.co/40sQCu9

Additional Show Notes

Episode Links:

• Nicholas' LinkedIn https://www.linkedin.com/in/nlarusstone/

• Sphinx Bio https://www.sphinxbio.com/

• Bits in Bio https://www.bitsinbio.org/

• Folding Playground https://www.sphinxbio.com/post/folding-playground

• Google Co-Scientist Paper https://arxiv.org/abs/2502.18864

• Recursion https://www.recursion.com/

• Defining TechBio https://medium.com/cantos-ventures/what-is-techbio-ef01cd61834d

Episode Chapters:

• 00:00 - Preview & Introduction

• 04:34 - Why Biotech & Life Science

• 07:33 - About Small Molecule Discovery Companies Octant & Benevolent AI

• 11:47 - Why Is Software Second in Biotech?

• 14:35 - What Engineers Bring to Biotech

• 16:00 - Is Science Data-Starved or Data Rich?

• 19:23 - How Scientists Work With Data

• 28:10 - Observations That Led to Founding Sphinx Bio

• 31:48 - AI Co-Scientists

• 37:57 - How Platforms like Sphinx Are Powered

• 40:31 - Foundation Model Companies' Interest in Science

• 43:58 - How Sphinx Leverages AI Models

• 45:59 - Managing Hallucinations

• 47:52 - Scientists: Give AI Another Try

• 49:02 - The Business and Cycles of TechBio

• 53:32 - Google's Co-Scientist, Notebook LM, and Other Science Ventures

• 57:43 - Protein Folding Playground

• 00:58:58 - "Wrapper" Companies In Era of Foundation Model Commoditization

• 01:01:56 - Bits in Bio Community & Growth

• 01:04:25 - Try Sphinx's Platform; Work for Sphinx; Support Your Local Scientist

“We can have people constantly inserting new experiments, and the lab’s going to manage everything, calling us from time to time just to clean some things” – David Milsted, Lab Systems Engineer

Imagine you’re a VP at a top EV battery manufacturer. Your biggest clients – Mercedes, BMW, Ford, Rivian – are sounding the alarm: your lithium-ion batteries are prone to dangerous flare-ups. Even when they function safely, they degrade too fast, stoking range anxiety and dragging down the resale value of their cars.

The fix seems obvious: get back in the lab and reformulate your battery’s chemistry. But where do you even begin?

There are 90-some naturally occurring elements on the periodic table, and choosing just six from a pool of 30 viable ones gives you over half a million possible combinations. Which ones definitely won’t work? Which ones might? And how do you accelerate the trial-and-error process of testing new materials?

That’s why I was so excited to visit the “A-Lab” at Lawrence Berkeley National Labs, an autonomous facility for accelerating materials discovery managed for the US Department of Energy. With a mix of AI and robotics, the A-Lab is making it possible to explore large chemical spaces faster than ever, with an initial focus on developing safer, better batteries.

During my tour and conversation with David Milsted, a postdoctoral researcher at the lab, we explored:

• How automated labs predict, synthesize, and test materials

• The orchestration of autonomous experiments through the A-Lab OS

• The future of materials characterization using automated microscopy

• Behind-the-scenes insights into running complex autonomous lab systems (e.g. robotic dispensers, speed mixers, and automated furnaces)

• David’s personal journey into lab automation, starting with his wood science studies in Brazil 🇧🇷

Watch or Listen Now

If you’re excited by automation, materials science, and seeing how labs can drive innovative research around the clock, join us for this conversation and lab tour.

Also available on:

Spotify: https://spoti.fi/42Uvv5J
Apple Podcasts: https://apple.co/40sQCu9

Additional Show Notes

Episode Links:

• David's LinkedIn https://www.linkedin.com/in/dmilsted/

• Lawrence Berkeley National Lab https://www.lbl.gov/

• A-Lab https://newscenter.lbl.gov/2023/04/17/meet-the-autonomous-lab-of-the-future/

• Ceder Group https://ceder.berkeley.edu/research-areas/autonomous-experimentation-for-accelerated-materials-discovery/

• UC Berkeley https://www.berkeley.edu/

• Mississippi State University https://www.bioproducts.msstate.edu/

• Wood Science https://en.wikipedia.org/wiki/Wood_science

Episode Chapters:

• 00:00 - Preview & Introduction

• 04:37 - LBNL Fun Facts

• 05:40 - David’s Journey Into Lab Automation (Wood Science, Automation Engineering)

• 10:26 - Why Wood Science

• 14:54 - Journey From Brazil To LBNL

• 25:50 - Material Discovery Process Steps

• 29:37 - Predicting Materials Properties

• 35:07 - Synthesizing Materials With Automation

• 43:21 - Testing Novel Materials

• 44:58 - Orchestrating Automated Labs (A-Lab OS)

• 49:14 - Shifting From Independent to Team-Based Projects

• 53:28 - University-Government Lab Structure

• 57:41 - What Inspired A Lab’s Vision

• 01:02:40 - Growth in Autonomous Lab Investment

• 01:05:34 - Future of Characterization Via Microscopy

• 01:10:22 - A Lab Tour (Core Components)

• 01:15:44 - Vendor Support: Mitsubishi, Labman, MTI, etc

• 01:18:58 - Managing Input Inventory For Runs

• 01:20:12 - Furnace Deep Dive

• 01:24:06 - Parallel Processing

• 01:25:43 - Vendor Automation Out of the Box; Importance of Societal Awareness

“Stage your automation. Only automate what you really need to, and wait until you really need it.”

I love this insight from Transon Nguyen, CEO of Eigen Therapeutics, a Bay Area biotech startup leveraging lab automation and AI to accelerate the development of cancer priming therapies.

One thing that struck me about Transon’s lab is how it was designed to be easily reconfigured to adapt to new research needs. Accomplished in part through putting wheels on carts, and creating islands for manual-work, walk-up instruments, and full automation.

In my lab tour and conversation with Transon (and guest co-host Jimmy Sastra of Monomer Bio!), we explored:

• How Eigen Therapeutics designed their lab for speed and flexibility

• Eigen’s custom-built LIMS and automated workflow

• Why priming is crucial for antibody-drug conjugate (ADC) and radioligand cancer therapies

• Why empathy is so critical to their company culture

• The value of emerging AI foundation models

Watch or Listen Now

If you’re curious about how cutting-edge biotech startups are really using AI and lab automation, and what it takes to design for speed, check out the full episode:

Also available on:

Spotify: https://spoti.fi/42Uvv5J
Apple Podcasts: https://apple.co/40sQCu9

Additional Show Notes

Episode Links:

• Eigen Therapeutics https://www.eigentx.com/

• Transon Nguyen's LinkedIn https://www.linkedin.com/in/transonn/

• Monomer Bio https://www.monomerbio.com/

• Jimmy Sastra's LinkedIn https://www.linkedin.com/in/jimmy-sastra-phd-11a0521/

• Radioligand Therapy https://www.novartis.com/research-and-development/technology-platforms/radioligand-therapy

• Antibody–Drug Conjugates (ADCs) https://en.wikipedia.org/wiki/Antibody%E2%80%93drug_conjugate

• Laboratory Information Management System (LIMS) https://en.wikipedia.org/wiki/Laboratory_information_management_system

Episode Chapters:

• 00:36 - Introducing My Guest Co-Host - Jimmy Sastra of Monomer Bio

• 11:24 - Introducing Transon Nguyen, CEO of Eigen Therapeutics

• 12:54 - Transon's Path Into Biomedical Engineering

• 15:45 - Science In Industry vs Academia - Importance of Culture

• 18:27 - Co-Founding Eigen Tx - A Confluence of Factors

• 24:27 - How to Find New Ideas

• 25:15 - Cost of ADCs and Radioligand Therapies

• 27:00 - Testing The Effectiveness of Priming Therapies

• 31:23 - How Patients Take Priming Therapies

• 33:19 - Importance of Empathy In Multidisciplinary Teams

• 37:54 - Lab Tour - Intro, Flexible Design

• 39:33 - Different "Homes" / Areas of The Lab

• 43:32 - Custom Built LIMS

• 48:54 - Robot Test Suite Demo

• 51:41 - A Look At Custom-Built LIMS

• 55:56 - Favorite Instruments

• 57:44 - Business of BioTech - Making The Numbers Work

• 59:09 - Don't Over-Automate

• 1:01:13 - Keeping The Lab Clean

• 1:03:01 - Moving Beyond Just Solving Problems

• 1:05:05 - Reframing Personal Identity and Seeing the Whole System

• 1:08:31 - Relevance of AI Foundation Models, Virtual Cells to Eigen's Mission

• 1:12:50 - Contact Transon to Learn More About Priming Therapies

“There’s a human component in every part of this job.”

I love this insight from Sudha Gollapudi on the empathy and communication skills needed to unlock the full potential of scientific lab automation.

Automating parts – or all – of the research and development process is a key strategy for accelerating scientific discovery. And, limits to current technology, challenges with integrating equipment from different vendors, and even cultural barriers make it easier said than done.

To help me make sense of what it will take to accelerate scientific discovery through lab automation, I recently sat down with Sudha Gollapudi and Madeline Wolf of The Automated Lab, a resource hub and contract engineering service for lab automation. Sudha and Madeline cut their teeth in lab automation at innovative organizations such as Genentech, Ginkgo Bioworks, and Calico, a subsidiary of Google’s Alphabet.

In our wide-ranging conversation, we explored:

• What lab automation actually encompasses (with examples)

• Modular-first design in lab automation

• Solving the human-robot interface with NLP & flexible data infrastructure

• Adapting scientific culture to increased automation

• Hot takes from the recent SLAS lab automation and screening conference

• Sudha and Madeline’s career journeys into lab automation

• Lab automation’s environmental impact

Watch or Listen Now

If you want to explore the potential of lab automation – or get a solid introduction into the space from two sharp experts – catch out our full conversation:

Also available on:

Spotify: https://spoti.fi/42Uvv5J
Apple Podcasts: https://apple.co/40sQCu9

Additional Show Notes

Episode Links:

• The Automated Lab: https://theautomatedlab.com/

• Open Source Resource Page: https://theautomatedlab.com/article.html?content=open-source

• Vivo Hamilton Liquid Handler Chatbot: https://hamilton-bot.sudhagollapudi.com/

• Automated Lab Patreon: https://www.patreon.com/theautomatedlab

• Sudha's LinkedIn: https://www.linkedin.com/in/sudha-gollapudi-21319029/

• Madeline's LinkedIn: https://www.linkedin.com/in/madelinewolf/

• ELISA Biochemistry Assay: https://en.wikipedia.org/wiki/ELISA

• Society for Laboratory Automation and Screening (SLAS): https://www.slas.org/

Episode Chapters:

• 00:54: Episode Introduction & Preview

• 05:38: Sudha and Madeline’s Academic + Professional Experience

• 06:53: Madeline’s Path Into Biotech Lab Automation

• 09:40: Sudha’s Path Into Biotech Lab Automation

• 10:55: Career Pathways Into Lab Automation

• 15:38: What Encompasses Lab Automation?

• 17:29: Lab Automation Example: Workcell for ELISA Immunological Assay

• 22:19: Environmental Impact of Lab Automation

• 27:31: SLAS Conference Hot Takes; Move to Smaller Footprint Products

• 36:28: Modular-First Design In Lab Automation

• 41:51: LIMs, ELNs, And Making Data More Accessible

• 46:10: Solving the Human-Robot Interface With NLP, Flexible Data Infrastructure

• 52:23: Automation For R&D vs Manufacturing

• 57:07: Adapting The Culture of Science To Increased Automation

• 1:00:29: Standardizing Scientific Language

• 1:08:02: How To Partner With The Automated Lab

Last week, the field of AI for life science hit a milestone with the release of Evo 2: the largest AI model for biology to date, trained on over 9 trillion DNA tokens from over 100,000 species. Evo2 is a generalist model that can, for example, predict whether a specific genetic mutation might cause a disease, such as cancer.

This model was developed through a partnership between NVIDIA and the Arc Institute - an innovative nonprofit research center.

As the Arc team was preparing for Evo 2’s release, I had the opportunity to sit down with Hani Goodarzi, a Core Investigator at the Arc Institute, quantitative and computational biologist, associate professor at UCSF, and co-founder of two startups, one of which – Exai Bio – is focused on the early detection of cancer.

A fun anecdote: when OpenAI co-founder Greg Brockman took a sabbatical last year, he spent part of that time collaborating with Hani and the Arc team to help them overcome the engineering challenges that come with scaling a biology foundation model of this size.

In my podcast interview with Hani we explored:

• The Emerging Promise of Biology Foundation Models Such as Evo 2

• Hani's Experience Partnering with NVIDIA and Greg Brockman of OpenAI

• Building Multidisciplinary Teams to Tackle Grand Challenges In Life Science

• The Science Behind The Early Cancer Detection Platform Developed by Exai Bio

• How Startups Help Bridge the Gap Between Academia and Industry Scale

Watch or Listen Now

If you're interested to get a peek into the hopeful future of early cancer detection and treatment through the use of powerful AI models, join us for this conversation:

Also available on:

Spotify:

Apple Podcasts: https://apple.co/40sQCu9

Episode Buzz

We were happy to see Patrick Collison, co-founder of Stripe and the Arc Institute, re-share our episode with Hani:

More community reactions:

Show Notes

Episode Links:

• Arc Institute https://arcinstitute.org/

• Goodarzi Lab https://arcinstitute.org/labs/goodarzilab

• Exai Bio https://www.exai.bio/

• Vevo Therapeutics https://www.vevo.ai/

• UCSF Cancer Center https://cancer.ucsf.edu/people/goodarzi.hani

• Hani's LinkedIn https://www.linkedin.com/in/hgoodarzi/

• Greg Brockman's Post https://x.com/gdb/status/1864052838256808339?lang=en

• NCI Cancer Stats https://www.cancer.gov/about-cancer/understanding/statistics

Chapters:

• 00:00 - Teaser

• 02:12 - Episode Introduction & Preview

• 07:35 - Hani’s Path Into Quantitative and Computational Biology

• 11:55 - The Emerging Promise of Foundation Models Such as Evo 2

• 15:53 - Dry Side vs. Wet Side of Biology

• 22:23 - Pivoting Into AI for Life Science From Unrelated Domains, and the Importance of Culture in Cross-Functional Teams

• 27:28 - How to Develop Multidisciplinary Teams: Brutal Honesty

• 30:07 - How Arc Institute’s Org and Funding Structure Enables Innovative Science

• 36:24 - How Tech Founders Are Pushing Biotech Innovation

• 39:07 - Hani’s Collaboration with NVIDIA and Greg Brockman of OpenAI

• 43:57 - The Science Behind Exai’s Early Cancer Detection Platform

• 52:05 - DNA Methylation to Improve Cancer-Specific Signals From Blood

• 55:25 - UCSF’s Application of This Technology for Breast Cancer Detection

• 58:34 - How Startups Help Bridge the Science “Valley of Death”

• 01:04:24 - Supporting Hani’s Work Through Scientific Contributions, Evangelism, and Applying the Scientific Method Across Disciplines

New Name, Even Better Content. As we launch into 2025, we’re bringing our “AI for Science” newsletter and podcast under one unified brand: Discovery Engines. Expect the same AI for Science insights you love, now paired with in-depth interviews that go beyond the headlines, available for you on YouTube, Spotify, and Apple Podcasts.

Sponsorships: Activated. This newsletter and podcast are a labor of love. Your support helps us cover essentials like studio rentals (for recording interviews), A/V equipment (to keep things polished), and travel (to meet and recruit unique guests). Huge thanks to those who’ve already pledged their support. Support us here with a paid subscription. Is your company or startup interested in sponsoring too? Drop us a line.

Alrighty, enough housekeeping. Let’s dig into this month’s latest in accelerating scientific discovery. –Nabil

• Arc Institute & NVIDIA Partner to Accelerate Computational Biomedical Research: Arc Institute, a research organization rethinking how science is done, has announced a new partnership with NVIDIA to scale AI-driven biological research. Backed in part by Stripe Co-Founder Patrick Collison, Arc is exploring new ways AI can design, predict, and understand biology. This collaboration builds on Arc’s previous work with Evo, an AI model built to predict and design at the level of DNA and across RNA and proteins in single-celled organisms.

• OpenAI’s “GPT-4b micro” Targets Longevity Science: In another sign of large AI labs moving into the natural sciences, OpenAI says it trained a new AI model, in partnership with Retro Biosciences, to re-engineer proteins that turn human skin cells into stem cells. One impact? To contribute to the field of longevity, which aims to increase healthy lifespan. The model is still a custom, private model. More coverage via the MIT Tech Review, here.

• NSF Funding Faces Major Cuts Under New Administration: The U.S. National Science Foundation was established in 1950 by Congress to promote the progress of science, advance national health and secure national defense. According to recent reporting, budget cuts and a change in directives on what science gets funded are headed to the agency. According to recent reporting, the NSF’s funding could drop from $9 billion to $3 billion, with up to 50% staff reductions possible. Additional coverage here.

• AI Increases Research Productivity, But At What Cost? Andrew Maynard covers a recently-published study from MIT’s Aidan Toner-Rodgers which “shows a significant increase in research and development productivity when scientists used an AI tool designed specifically to increase the rate of discovery of new materials.” However, “while enjoyment from increased productivity partially offsets this negative effect – especially for high-ability scientists – 82% of researchers see an overall decline in satisfaction.” Original study linked here.

Subscribe now

"...there’s tenacity, curiosity, hard work, brain power, but then there’s also: luck, timing and mentorship. Which I think are so important in the ultimate success that you achieve as a researcher."

I love this insight from Hemai Parthasarathy, a neuroscientist (Johns Hopkins, MIT), experienced editor (Nature, PLOS), and expert in helping deep tech startups scale (Thiel Foundation, Google X, MIT The Engine).

Recently, I sat down with Hemai for a discussion on my podcast: Discovery Engines. In our conversation, we explored:

• Hemai's journey from the lab bench to scientific publishing and deep tech acceleration.
• How scientific publishing has adapted to new trends in technology and open science.
• The founding story and model behind the Thiel Foundation's Breakout Labs.
• Navigating the political headwinds impacting the deep tech funding landscape.
• The key factors and the importance of patient capital for deep tech startup success.

If you're interested to hear how scientist founders make it from discovery to market success, join me in this deep dive with Hemai as she shares her unique experience across the deep tech ecosystem.

Watch/Listen now: YouTube | Spotify Podcasts | Apple Podcasts

• Predicting 3D Genomic Structure from DNA Sequences: MIT chemists have developed ChromoGen, a generative AI model that predicts how a specific DNA sequence will arrange itself in the cell nucleus. Their model can predict thousands of structures in minutes, enabling researchers to more easily study how the 3D organization of the genome affects individual cells’ gene expression patterns and functions. Data and model available here.

• AMD & Johns Hopkins Launch LLM-Powered Agent Lab: While some of the more headline-grabbing AI for Science efforts are focused on areas such as building biological foundation models, some might argue that automating the remainder of the scientific development process is equally impactful. AMD and John Hopkins’ Agent Laboratory: “consists of specialized agents driven by large language models to support you through the entire research workflow – from conducting literature reviews and formulating plans to executing experiments and writing comprehensive reports.” Try it here.

• 2024 State of Techbio Survey: Key Insights: TechBio community organization Bits In Bio recently published survey results from 200+ respondents, who shared how they’re leveraging emerging tools to push boundaries in biotech. What stood out for me:

  • ML Models: "60% of respondents use proprietary or in-house models regularly, surpassing GPT-4 (52%) and AlphaFold (34%)

  • ML Applications: "30% of respondents prioritize library/in silico screening to accelerate drug discovery, 29% focus on target identification/selection for therapeutic development, and 25% apply ML for lab automation.”

  • Lab Automation: "Over 60% of respondents use lab automation for data acquisition and analysis, while 58% employ it for liquid handling, pipetting, and sample preparation/extraction." Download the full dataset here.

• The Block Universe: Everything All At Once. Some interpretations of Einstein’s theory of relativity suggest that all of time – past, present, and future – exists simultaneously, much like a movie encoded onto a DVD. Our lived experience might just be our consciousness moving through specific “frames”—similar to how a laser scans a disc or a needle traces the grooves of a vinyl record. Mind-bending? Absolutely. And it raises profound questions about free will and the nature of time itself. For a fun, visual explainer, check out this video from the YouTube channel Kurzgesagt:

Find this newsletter valuable? Subscribe for regular insights, and share it with friends who are passionate about accelerating scientific discovery.

Subscribe now

Help us improve. Like a good neural network, this newsletter is only as effective as the input it receives. We’d love to hear your feedback – let us know how we can make it even better for you.

Share feedback

Catch you on the next one 🤙🙌 – Nabil

"...there’s tenacity, curiosity, hard work, brain power, but then there’s also: luck, timing and mentorship. Which I think are so important in the ultimate success that you achieve as a researcher."

I love this insight from Hemai Parthasarathy, a neuroscientist (Johns Hopkins, MIT), experienced editor (Nature, PLOS), and expert in helping deep tech startups scale (Thiel Foundation, Google X, MIT The Engine).

Recently, I sat down with Hemai for a discussion on my podcast: Discovery Engines. In our conversation, we explored:

• Hemai's journey from the lab bench to scientific publishing and deep tech acceleration.
• How scientific publishing has adapted to new trends in technology and open science.
• The founding story and model behind the Thiel Foundation's Breakout Labs.
• Navigating the political headwinds impacting the deep tech funding landscape.
• The key factors and the importance of patient capital for deep tech startup success.

If you're interested to hear how scientist founders make it from discovery to market success, join me in this deep dive with Hemai as she shares her unique experience across the deep tech ecosystem.

Catch the full conversation here:

Also available on:
Spotify: https://spoti.fi/42Uvv5J
Apple Podcasts: https://apple.co/40sQCu9

“You sit in between three of the biggest problems in the [biotech] industry.”

I love this insight from Luis Villa – co-founder of Bay Area Lab Automators – on why a career in biotech lab automation is so compelling, and impactful.

Recently, I sat down with Luis on my podcast: Discovery Engines. In our conversation, we explored:
• A day in the life of an automation engineer at a cell therapy manufacturing startup
• The state of the lab automation market today
• The rise of closed-loop/autonomous labs transforming science
• Recent pushes to onshore life science research and manufacturing
• Practical ways to get involved in the Lab Automation community

If you're interested in the intersection of engineering, biotech, and automation, this episode is for you.

Catch the full conversation here:

Also available on:
Spotify: https://spoti.fi/42Uvv5J
Apple Podcasts: https://apple.co/40sQCu9

• DeepMind Launches GraphCast AI Weather Model: According to a recently published paper in Nature (“Probabilistic weather forecasting with machine learning”), Google DeepMind sidestepped traditional physics-based weather simulation techniques to develop a highly accurate ML-based model for predicting 15-day global forecasts. Google claims their approach delivers forecasts more accurately and more quickly than gold-standard weather simulation systems, a feature especially important given natural disasters such as the wind-fueled Southern California wildfires. (PSA: Having previously volunteered for the Red Cross’ wildfire relief efforts in LA, I can attest to the excellent work they do. Click here to donate the Red Cross’ relief fund.)

• E11 Launches PRISM Connectomics Platform: Non-profit startup E11 Bio announced PRISM, or “Photo-connectomic Reconstruction by Iterative Staining with Molecular annotations.” This new platform integrates genetic labeling, microscopy, and AI modelling to more affordably model and map mammalian brains. The goal? Advancing our understanding of brain disorders and inspiring the development of brain-inspired AI systems. More info here.

• US and China Renew Science Pact: In a time when there is increased pressure for the US to “decouple” from China and other states, a long-running science pact between the US and China was renewed for another five years. This deal “sustains cooperation between government and university researchers in basic science such as earthquake and health monitoring.” Read more in the Wall Street Journal (paywall).

• Andreessen Horowitz + Lily Biotech Launch Ecosystem Venture Fund: Lilly and a16z team up to deploy a $500 million fund focused on “advancing the development of new medicines, enabling novel modality platforms, and scaling emerging health technologies.”

Subscribe now

Our North Star is to accelerate scientific discovery. One way we believe we can help is by shining a light on the founders and innovators who are boldly transforming how science is done, and helping them build the partnerships, hire the talent, and close the deals necessary to get their ideas out of the lab and into the world.

That’s why we’re thrilled to introduce Discovery Engines – our podcast dedicated to these changemakers. As your host, I (Nabil) will bring you inspiring stories and actionable insights from trailblazers in science and technology, starting right here in our backyard in San Francisco.

In our debut episode, I sit down with Cristian Ponce and Théo Schäfer, co-founders of Tetsuwan, a startup pioneering AI scientists to accelerate life science research. We delve into their founding journey, their first customer deployment, the state of life science robots, rules-based vs. machine learning-based AI, and the future of lab automation.

We hope you enjoy listening as much as we enjoyed creating this episode. If you do, consider subscribing on your favorite platform for more inspired content.

Watch now: YouTube | Spotify Podcasts | Apple Podcasts (audio only)

• The Well and Multimodal Universe Datasets: Research collaboratives Polymathic AI and Multimodal Universe released 115 terabytes (!) of validated scientific data in a variety of domains including fluid dynamics, biological pattern formation, acoustic wave propagation, and astronomy. The purpose of such a diverse dataset is to help AI exploit common phenomena across sciences, such a waves. More context here from Miles Cranmer.

• Bluesky Starter Pack: As the science community increasingly embraces the social media platform Bluesky, this crowdsourced effort attempts to help us wayfind. With a heavy emphasis on Chemistry, it also includes lists of experts and organizations in Science Communication, Open Source Machine Learning, and Research Policy, among others.

• Elman R&D Platform: London based Venture Studio Deep Science Ventures released Elman, a platform leveraging LLM agents to offer workflows for R&D, “from initial approach design to final development.” Sign up to try it here.

• CERN Particle Physics Course: The European lab for particle physics, and home of the Large Hadron Collider, now offers a free online course to learn about everything from Higgs boson to antimatter. Duration of entire video library is four hours; direct link here.

• Is Everything Conscious? As AI engineers, ethicists, and others explore the definition of consciousness, and the ethical questions that arise with creating simulated conscious agents, some argue that it’s a moot point: from birds to rocks, the universe is already inherently teeming with consciousness. Rooted in Giulio Tononi’s Integrated Information Theory, this article in Quartz lays out this thought-provoking hypothesis.

  

Find this newsletter valuable? Subscribe for regular insights, and share it with friends who are passionate about AI for Science.

Subscribe now

Help us improve. Like a good neural network, this newsletter is only as effective as the input it receives. We’d love to hear your feedback – let us know how we can make it even better for you.

Share feedback

Catch you on the next one 🤙🙌 – Nabil

Animal studies are a cornerstone of biomedical research, yet many of the tools used to conduct them haven’t changed in decades. Michael Kaca, Co-Founder and Chief Science Officer of Olden Labs, is on a mission to change that. In this episode of Discovery Engines, we explore how his team is modernizing animal research with custom hardware and machine learning-powered analytics: bringing rigor, efficiency, and automation to an essential but often overlooked part of life sciences.

In our chat we covered:

• Why animal studies remain essential in biomedical research

• How artificial organs and AI models could reduce reliance on animal testing

• Olden Labs’ journey from an idea to a hardware & AI startup

• What happens when a roboticist enters the life sciences

• The DOME Smart Cage – a real-time tracking system powered by ML

• How better data can lead to breakthroughs in genetics and longevity

Listen Now

If you’re interested in the intersection of AI, hardware, and the future of life sciences, you’ll be interested to hear Michael’s insights on how technology can enable data-driven, ethical, and scalable scientific research.

Catch the full conversation here:

Also available on:
Spotify: https://spoti.fi/42Uvv5J
Apple Podcasts: https://apple.co/40sQCu9

The future of scientific discovery could look very different if AI-powered wet-lab scientists were leading the charge. Imagine automated systems designing, executing, and iterating on experiments at a scale and speed far beyond human capability. That’s the vision behind Tetsuwan, an early-stage startup building the next generation of autonomous lab AI.

On this episode of Discovery Engines, I sit down with Cristian Ponce and Théo Schäfer, Co-founders of Tetsuwan, fresh out of stealth mode following their first client delivery. We dove deep into what it takes to bring AI-driven lab automation to life and explored some of the biggest challenges and opportunities in this space.

Specifically, we covered:

• How large language models are transforming lab automation by translating scientific intent into executable code

• The challenge of scientific language and biased data, and how to overcome it

• How AI can increase transparency and collaboration in lab automation

• The differences between launching a deep tech company in the US vs. Europe

• The safety implications of AI-driven wet labs and how to manage risk

• Why Astro Boy is the unofficial inspiration for their startup

A Tsunami Warning (Literally)

As we sat down to record in downtown San Francisco, our phones went off in unison with a tsunami warning. Fortunately, the tsunami never came, but Cristian and Theo stayed as unflappable as one might expect from founders building a disruptive deep tech company.

Listen Now

If you’re curious about how AI and robotics are reshaping the wet lab, this episode is for you. Catch the full conversation here:

Also available on:
Spotify: https://spoti.fi/42Uvv5J
Apple Podcasts: https://apple.co/40sQCu9

In her talk, “Developing Machine Learning Methods at Scale for Chemistry, Materials, and Climate Problems,” Aditi Krishnapriyan, Assistant Professor at UC Berkeley in Chemical Engineering and EECS, outlined her group’s dual focus:

1. Developing fundamental ML models for chemistry and materials problems

2. Using scientific insights to create new ML methods that don’t yet exist

According to Aditi, at the core of chemical and materials research is the potential energy surface – a representation of the energy landscape that governs molecular interactions and reactions. This surface is critical for understanding processes such as how molecules bind to a protein.

However, calculating these surfaces via quantum mechanics is computationally expensive and doesn’t scale well. Experimental methods are often prohibitive. ML has become a powerful surrogate, offering faster alternatives to these resource-intensive simulations.

Aditi posed a key question: How do we develop ML methods that generalize across applications while leveraging modern computing capabilities? Many systems studied in science, such as molecules and climate models, can be represented as graphs. Inspired by the now landmark paper “Attention Is All You Need,” Aditi and her team explored how attention mechanisms could be applied to graph-based systems.

Aditi and her team developed a new graph attention mechanism for molecular representations, achieving top performance across benchmarks in catalysis, materials, and molecular applications. Remarkably, they were the only academic team to outperform industry giants like Meta and Microsoft on datasets like Open Catalyst. Aditi sees this as a sign of increasing democratization in the AI for Science field, bolstered by open science collaboration. To support this, her team made all their code and models publicly available.

Why This Matters

According to Aditi, one impactful application of these ML models is in identifying transition states, which represent the lowest energy pathways between two molecular states. Understanding these pathways is crucial in many areas, including rechargeable batteries. Transition states are key to processes like electrolyte decomposition, which directly affects energy density, lifespan, and safety in batteries.

Despite these successes, challenges remain. For example, while amusing failures in text-generation AI can be fixed easily, scientific ML models demand greater reliability due to the critical nature of their applications. Aditi’s team is addressing this by exploring multi-modal training strategies that incorporate experimental data alongside simulations, yielding promising results.

Additionally, scientists often prefer faster, specialized models for specific applications (e.g., amino acids or perovskites), driven by GPU scarcity and their domain-specific needs. To address this, Aditi’s team is working on distilling large models into smaller, more efficient versions tailored for specific use cases.

For Aditi’s full talk, and additional insights into cutting-edge AI for chemistry and materials research from experts Wen Jie Ong (NVIDIA), Arash Khajeh (Toyota Research Institute), and Aldair Gongora (Lawrence Livermore National Lab), check out our full recording:

Subscribe now

• OpenAI, Sanofi, and Formation Bio announce Muse, an AI-powered tool designed to improve drug development by speeding up clinical trial recruitment.

• Google.org announces $20 million fund to support AI for scientific breakthroughs. Press release calls out fields such as rare and neglected disease research, experimental biology, materials science, and sustainability.

• AI drug discovery company Recursion completes its acquisition of Exscientia, moving Recursion toward more vertically-integrated and technology-enabled drug discovery.

• Arcadia Science publishes SwimTracker, a high-throughput imaging approach to track and quantify single-cell swimming.

• Venture studio Flagship Pioneering seeks a Director of Product to “set the roadmap for the autonomous materials lab of the future and collaborate across disciplines.”

• UCSF’s Hsiung Lab seeks multiple postdocs to contribute to their research in synthetic gene regulation, combinatorial genetics, and tissue biology.

• DeepMind spin-off Isomorphic Labs seeks candidates for multiple open roles to meet their hiring team at NeurIPS.

• Vancouver (NeurIPS Conference), Dec 11: “Breaking Silos” social, an open community gathering for AI x Science, where they will explore how open and collaborative AI is transforming scientific discovery.

• SF Bay Area, Dec 11: AI + Health Demo Night, exploring where ML models are finding their way into clinical workflows, biosensors, and healthtech.

• SF Bay Area, Dec 12: Laboratory Robotics Interest Group (LRIG) End of Year Event, featuring networking and technical presentations by Octant.

How are LLMs doing at answering PhD-level science questions? A fun graph courtesy of physicist and cosmologist Anthony Aguirre.

Find this newsletter valuable? Subscribe for regular insights, and share it with friends who are passionate about AI for Science!

Subscribe now

RLHF! Like a good neural network, this newsletter is only as effective as the input we receive from you. We’d love your feedback—drop us a note on how we can make it better for you.

Share feedback

Catch you on the next one! – Nabil

Last month, Dario Amodei, CEO of the major AI lab Anthropic, penned an in-depth post on the upsides of continued AI development. An early member of the OpenAI team, Dario co-founded Anthropic with his sister Daniela in part to more seriously research AI safety and alignment. His post – Dario explains – serves as a counterweight to the risk-focused conversations he and Anthropic prioritize.

The first area Dario called out? Biology and health.  Amodei takes pains to make a distinction between AI models that conduct better data analysis, and AI scientists who perform all of the tasks that biologists do – including controlling labs or giving instruction to scientists – much in the same way a PI might direct their graduate students.

Amodei – once a postdoc at Stanford University School of Medicine – writes that much of our progress in biology has come from a handful of fundamental breakthroughs, such as CRISPR for gene editing, genomic sequencing and synthesis, optogenetic techniques for firing neurons, and mRNA vaccines.  “There’s perhaps ~1 of these major discoveries per year and collectively they arguably drive >50% of progress in biology.” 

Further, Dario writes that many of these discoveries didn’t hinge on new technology; rather, they were waiting in the wings, until a scientist connected insights from various fields. An AI biologist, he claims, would be effective at precisely these sorts of discoveries, contributing a foundational set of technologies upon which the rest of modern life science could accelerate its work.

Why This Matters

Major AI labs such as OpenAI and Anthropic have gone all-in on creating productized enterprise platforms to provide end-user capabilities in areas such as developing better marketing copy, market analysis, and integrated in-app chat-bots. The strategy, in some ways, make sense.  These models – trained on large text-based datasets – lend themselves more easily to these applications, which themselves are easier to package and scale for revenue growth.  Important considerations when AI talent and compute are exceedingly expensive. 

That said, it's also clear that scientific discovery lies close to the hearts of their chief executives.  When asked what insights he would be most excited for an AGI system to provide, OpenAI’s CEO shared: a grand unified theory of physics. In their most recent “o1” model launch, OpenAI showcased the way their model helped Catherine Brownstein, a geneticist, conduct her research more effectively. Last year, news organizations claimed that a major reason for Amazon’s $4B investment in Anthropic was to strengthen Amazon’s own platform for drug development, Bedrock. 

Google’s DeepMind has been playing in this field for over a decade, recently winning the Nobel Prize in Chemistry for its protein-folding prediction model AlphaFold, and even spinning out Isomorphic Labs, a startup focused on drug discovery.  Could this post from Amodei, and recent science-focused marketing clips from OpenAI, signal an increasing focus on joining Google DeepMind in developing and productizing models to accelerate scientific discovery? Would these major AI labs become partners or competitors to existing players, especially in the biotech sector which has been applying AI for life sciences for the better part of a decade? We’re eager to watch where this goes.

Subscribe now

• Roei Herzig and team share RoboPrompt, a framework that enables off-the-shelf text-only LLMs to directly predict robot actions. #autonomouslabs

• Insilico Medicine – two years into their partnership with Sanofi – announces an AI-facilitated lead with first-in-class (FIC) potential against an undruggable transcription factor target for treating oncology diseases.

• Siemens augments its mechanical and electromagnetic simulation capabilities by purchasing US industrial software make Altair for $10B.

• Quanta Magazine article discusses an exciting proposal to detect gravitons – hypothetical particles thought to carry the force of gravity. #physics

• MIT’s Schwarzman College of Computing announces the Tayebati Postdoctoral Fellowship to support researchers accelerating AI for Materials Science and Engineering, among other science domains.

• The Meta FAIR team is seeking researchers interns for their AI for Chemistry work.

• The NYU Center for Data Science (CDS) seeks applicants for a Faculty Fellow position aimed at working at the boundaries between the data science and domain sciences such as biology, physics and chemistry.

• SF Bay Area, November 14: FutureHouse, the autonomous science nonprofit, hosts experts from Berkeley's Autonomous Labs of the Future among others for a talk, demo and pitch night.  Tickets here.

• SF Bay Area, November 20: The MIT Club of Northern California host experts from NVIDIA, Toyota, the DOE and UC Berkeley to explore recent efforts in AI for Autonomous Science. Tickets here.

• Online (Zoom), November 22: The NSF’s Institute for AI & Fundamental Interactions hosts Yuan-Sen Ting, Associate Professor in Astrophysics from Ohio State for a Public Colloquium on Expediting Astronomical Discovery with Large Language Models. Watch live here.

What happens when physicist Philip Moriarty of the University of Nottingham takes an early version of ChatGPT out for a spin? Lots of quirks and features. Take a look:

Find this newsletter useful? Subscribe for regular insights, and share it with friends who are passionate about AI for Science!

Subscribe now

RLHF! Like a good neural network, this newsletter is only as effective as the data it’s trained on. We’d love your feedback—drop us a note on how we can make it better for you.

Share feedback

See you next week. –Nabil 🙏🙌