The Bitter Lesson for Biology — Adam Green on Scaling Laws for Virtual Cells
In this episode, the founder of Markov Biosciences, @adamlewisgreen, explains the "bitter lesson" for biology, the idea borrowed from Richard Sutton that large unbiased datasets and the right training objective tend to outcompete models with hard-coded rules and human priors.
We talk a lot about virtual cells, sources of bias in data, and his evidence that virtual cells pre-trained on plain observational data show clean scaling laws, getting monotonically better at predicting unseen perturbations as the models grow, and beating a state-of-the-art model built specifically for that task.
Our interview was about 3 hours long, but cut to 1.5 hours. My main goal was to understand Adam's perspective on virtual cells and his goal to "solve biology."
Search for "The New Biology" on YouTube, Spotify, and Apple Podcasts.
Timestamps:
00:00 — Cold open
01:58 — Clinical predictions from a virtual cell
05:38 — What is a "virtual cell"?
08:01 — The problems with single-cell RNA-seq
11:31 — The urns analogy
19:54 — Observational vs. perturbation
23:29 — The bitter lesson for biology
29:06 — Geometric Plackett-Luce
38:27 — Ablations and loss function
47:23 — Cells as specimens
59:26 — Antibody-Drug conjugates
1:11:16 — Will we ever understand the cell fully?
Human biology matters. Scientists and AI need human data to understand health and disease.
Crownlands is open sourcing Gateway 4M, the largest single-cell tissue dataset ever released from living humans, to advance research on brain aging and neurodegeneration.
Integrated single-cell and spatial transcriptomic profiling in ALS uncovers peripheral-to-central immune infiltration and reprogramming: https://t.co/fIFqMcZKlz
🚨What if we could reliably program macrophage polarization state?🚨
https://t.co/6qGxuHptQq
Macrophages are highly plastic immune cells that perform critical functions by polarizing into distinct cellular states. The polarization state of macrophages can substantially influences the progression of cancers, infections, and autoimmunity. (1/11)
Probably one of the handful of companies in the space of "virtual cells" that IMO has the right strategies, data modalities and models to really show the power of what happens when you deeply couple expt design, ML models to well thought out questions & applications.
A @Nature study from Rockefeller's @junyue_cao describes a new platform called PerturbFate that reveals how diverse genetic perturbations funnel into shared disease states, a method that could unlock therapeutic targets for complex diseases.
🔗: https://t.co/hkkTbpvQkZ
@anshulkundaje articulates something the AI-for-biology practitioners (or AI-for-science for that matter) need to hear more: we are far from a stage that scale alone solves biology. Deep domain expertise and principled interpretation (as opposed to cherry-picking of results) is how we actually make progress. There's too much hubris right now in assuming one can brute-force their way through biological complexity without understanding it.
What a joy to be friends with such an exceptional scientist. Please read this elegant, simple, and groundbreaking article by @kevinguttenplan and the Freeman lab! It is worth your time.
https://t.co/Zla4YxZ4pa
In collaboration with Reuben Saunders, @JswLab, and Xiaowei Zhuang, we are very excited to release Perturb-Multi: a platform for pooled multimodal genetic screens in intact mammalian tissue.
Check it out!
https://t.co/iJ8hi3ddz4
AI x life sciences: Hype or hope or inevitability? Our firm (@LumaGroup_ like n follow!) take on this game-changing intersection, its impact, and VC's role in shaping the future. Read our whitepaper: https://t.co/WIonsIxvPG
Let me make it easy for those interested in the truth or who have been denying that these mobs are motivated by anti-Semitism or support for terrorism. Here is a short thread with just some of the direct evidence. I will just focus on the protests at and around Columbia: