Markov Biosciences, a startup in San Francisco, is betting that biology is about to have its GPT moment. In this episode, founder Adam Green 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. Adam thinks, in particular, that the virtual cell field took a wrong turn by spending hundreds of millions of dollars collecting expensive perturbation data. Green’s counterargument is that the data needed to train useful virtual cells is not limiting, but rather compute (and the loss function) are. By treating single-cell RNA-seq as a ranking problem rather than raw counts (a century-old idea traceable to a 1927 psychophysics paper), they found 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.

00:00 - Cold open and introduction 

01:58 - The first clinical prediction from a virtual cell

05:38 - What is a "virtual cell," really? 

08:01 - Single-cell RNA-seq biases and the urns analogy

23:29 - The bitter lesson for biology

30:55 - Geometric Plackett-Luce: the right loss function

59:26 Trop2 deep dive

1:11:16 - Top-down vs. bottom-up biology, mechinterp, and control as the goal 

Readings and mentions: 

• Markus Covert — A Whole-Cell Computational Model Predicts Phenotype from Genotype
• Markov's ADC-predictions thread (Adam Green)
• Scannell et al. (2012), "Diagnosing the decline in pharmaceutical R&D efficiency" (Eroom's Law)
• Adam Green on the Bitter Lesson
• Adam Green on RNA-seq issues
• Arc Institute — STATE model (Adduri et al., 2025)
• GPT-1: Radford et al. (2018), "Improving Language Understanding by Generative Pre-Training"
• Rich Sutton, "The Bitter Lesson" (2019)
• Yann LeCun's "cake" analogy (explainer)
• Markov paper — Generative ranking / Geometric Plackett–Luce (the GPL paper)
• Thurstone (1927), "A Law of Comparative Judgment"
• scBaseCount (Youngblut et al., 2025)
• CZ CELLxGENE Discover (data portal)
• X-Cell (Xaira Therapeutics), Wang et al. (2026)
• Adam Green / Markov, "A Future History of Biomedical Progress" (biocompute)
• Decoding TROP2 in breast cancer: significance, clinical implications, and therapeutic advancements
• Bunne et al. (2024), "How to build the virtual cell with artificial intelligence: Priorities and opportunities," Cell
• Nintil (2023), “Notes on end-to-end biology.”
  
  

Nonfiction Laboratories is building a technology called “magnetogenetics” that promises to control proteins inside the body — such as antibodies or enzymes — using small magnets. In this episode, co-founder Maria Ingaramo and scientific advisor Andrew York explain how they engineered a protein, MagLOV, that responds strongly to magnetic fields, why most prior attempts have failed to replicate, and how the mechanism of magnetically-controlled proteins actually works. They also get into the “dream” use cases, like cancer drugs that activate only at the tumor, which might have a lower toxicity inside the body. 

This podcast is made possible by Astera Institute.

Notes from our discussion: https://nikomc.com/essays/protein-magnets.html

00:00 - Opening

00:54 — Introduction

01:35 — The dream

05:38 — Why magnets vs. light or ultrasound

10:05 — The physics

17:48 — On the name "magnetogenetics"

21:25 — Birds and cryptochromes

27:09 — Why is the field filled with so much junk?

29:51 — Adam Cohen's molecule

33:24 — Markus Meister’s debunking

38:06 — The experiment

46:22 — Finding the LOV domain

54:11 — Singlets, triplets, and cysteine

56:54 — What the magnet is actually doing

1:05:13 — The conformational-change red herring

1:12:46 — The Quantum Biology Institute

1:19:31 — Founding Nonfiction Labs

1:24:38 — How to convince skeptical investors

1:29:39 — What a magnetogenetic medicine might look like

1:38:50 — First clinical indications

1:45:12 — The regulatory path

1:48:01 — What the field needs

1:54:30 — Appendix: Whiteboard lecture

Plasmidsaurus took plasmid sequencing from $600 to $15 and turned a "boring" service company idea into a hugely successful company serving 70,000+ scientists. In this episode, CEO Mark Budde and Niko McCarty get into the bigger question: what does it take for companies to automate and scale wet-lab biology methods in the same way that Plasmidsaurus did for sequencing? They cover the early Oxford Nanopore bet, the obsession with speed, and why Mark won’t sell customer data to AI labs. 

This podcast is made possible by Astera Institute.