Skip to content

Lab-in-the-Loop at Compute Speed — Ian Quigley, Leash Bio

Latent Space: The AI Engineer Podcast

1h 49m
Transcript

Original podcast

Where to listen

Summary

Written by a model from the transcript; not yet reviewed.

This episode explores how Leash Bio uses DNA-encoded libraries (DELs) combined with massive-scale data generation to train machine learning models for small molecule drug discovery. The guest, Ian Quigley, explains the historical trauma associated with DELs due to noise and false positives, and details how Leash Bio overcame these hurdles by screening millions of molecules across thousands of protein targets, including mutated variants. The discussion covers the failure of public datasets like ChEMBL due to model learning of chemist sociology rather than physics, the evaluation of models like Hermes, and the shift toward using natural language and reinforcement learning agents for drug discovery.

  • Screening massive combinatorial libraries against thousands of protein targets generates consistent, reproducible interaction data for machine learning. 17:26
  • Public chemical datasets often allow models to predict the creator rather than physics, learning chemist sociology instead of ground truth. 47:38
  • Engineered mutations in protein binding pockets provide necessary negative controls and causal activity cliffs for robust model training. 1:04:45
  • Using natural language representations and reinforcement learning agents enables rapid exploration of binding hypotheses against large experimental datasets. 1:17:34

Chapters

  1. 0:00
  2. 1:22
    Podcast Intro and Host Welcome

    The hosts introduce the episode and welcome the guest Ian Quigley.

  3. 2:24
    Leash Bio Founding Story

    Ian explains how Leash Bio was founded to address small molecule binding challenges.

  4. 4:00
    DNA‑Encoded Libraries Explained

    Ian details what DNA‑encoded libraries (Dells) are and how they work.

  5. 8:00
    Scaling with DNA Sequencing

    The team discusses the scalability and cost advantages of using DNA sequencing as a readout.

  6. 23:00
    Future Directions and Data Strategy

    The discussion turns to Leash’s strategy for scaling data collection and improving model performance.

  7. 32:18
    Kaggle Competition Design

    Ian describes how Leash designed a Kaggle competition to benchmark small‑molecule binding prediction.

  8. 40:38
    Hermes Model Development

    Ian outlines the creation of the Hermes model, a simple architecture combining protein and chemical embeddings.

  9. 42:25
    Attention Critical for Protein–Ligand Interaction

    The team discusses how scrambling positional coordinates in protein–ligand graphs shows attention mechanisms are essential for capturing interactions.

  10. 44:12
    Building Block Clustering to Reduce Memorization

    Introducing a scheme to cluster similar building blocks, preventing overlap between train and test sets to mitigate memorization effects.

  11. 45:26
    XGBoost Baselines Reveal Distributional Limits

    Comparing Hermes to XGBoost across split regimes highlights how far‑out‑of‑distribution data challenges simple models.

  12. 46:32
    Clever Hans: Chemist Identity Leakage

    A story of how models can predict chemist identity from molecular structure, exposing sociological biases in datasets.

  13. 51:15
    Using Chemist Probability Vectors for Activity Prediction

    Replacing chemical representations with creator probability vectors to avoid learning chemist fingerprints while maintaining predictive power.

Highlights

  • 4:06

    Scaling up to 40 trillion moleculesdata

    DNA-encoded libraries allow pooling millions to trillions of molecules in a single tube attached to DNA barcodes, achieving massive combinatorial scale compared to traditional single-well assays.

  • 27:33

    The four billion interaction Kaggle conteststory

    Leash Bio generated a 4 billion interaction dataset and hosted a Kaggle competition, only to discover that across the most challenging evaluation splits, top models performed no better than random chance.

  • 48:21

    Predicting the chemist from moleculesinsight

    Classifiers trained on ChEMBL data can predict which individual chemist created a molecule with 60 percent accuracy, demonstrating that models often learn human sociology rather than physical binding physics.

  • 52:20

    Molecular weight as a shortcutinsight

    In many public datasets, sorting by molecular weight serves as one of the single biggest predictors of binding due to the optimization habits of medicinal chemists.

  • 1:04:45

    Engineering protein pocket mutationsstrategy

    By physically mutating specific amino acids in protein binding pockets, researchers can create crucial negative controls to test whether models and molecules truly rely on intended physical interactions.

  • 1:15:56

    Using natural language for drug discoverystrategy

    Leash Bio employs open-source language models with natural language descriptions of small molecules, proteins, and activity cliffs to run rapid reinforcement learning rollout loops.

  • 1:45:36

    Detecting misfolded proteins with chaperonesstory

    When screening libraries, the appearance of molecular chaperone analogs indicates protein misfolding, providing a vital quality control filter for training data.

Quotes

  • “We downloaded all chemicals from all chemists at Kimball, which is the biggest repository of these things and their activities.”
    Unnamed speaker · 0:30
  • “Sequencing DNA is another way of taking pictures.”
    Unnamed speaker · 6:01
  • “The thing that we were doing, which are just basic smiles, kind of beat everything else.”
    Unnamed speaker · 33:52
  • “And that made us think that like attention was super critical.”
    Unnamed speaker · 42:36
  • “The models do not understand chemistry at all.”
    Unnamed speaker · 55:45
  • “We want this thing to really understand, like, these causal changes.”
    Unnamed speaker · 1:20:04
  • “And then just keep making choices.”
    Unnamed speaker · 1:23:48

About this episode

A few years ago I stumbled upon Leash Bio’s https://www.leash.bio/ Substack https://leashbio.substack.com/, and was hooked. Be it their Kaggle competition https://leashbio.substack.com/p/introducing-belka that showed that no, we really haven’t solved protein-ligand binding https://leashbio.substack.com/p/belka-results-suggest-computers-can, or their crazy discovery that knowing a molecule’s author is perhaps all you need to know to predict how good it is https://leashbio.substack.com/p/ai-for-chemistry-in-2025-is-like. Their substack — or at least Owl Posting https://www.owlposting.com/’s deep dive https://www.owlposting.com/p/an-ml-drug-discovery-startup-trying on them — should be required reading for any scientist trying to do chemistry with AI. So I was super excited to finally be able to sit down with Ian Quigley https://www.linkedin.com/in/iquigley/ from Leash Bio, and pick his brain. Leash specializes in DNA Encoded Libraries https://en.wikipedia.org/wiki/DNA-encoded_chemical_library (aka “DELs”), a way of collecting ridiculously high-throughput interaction data between proteins and small molecules. Before I discovered Leash, I had heard nothing but horror stories from DEL users. And yet, Ian and friends somehow found a way to tame the beast, extracting clean and reproducible data that is just perfect for large scale modeling. Without teasing too much, the answer comes down to epistemic rigor, and an incredible conviction that you just have to Do The Thing. It gets even better! Near the end of this episode, Ian teases something his team was just playing with when we recorded: an AI playground for agents to do verifiable scientific reasoning. They discovered the data they had spent the last few years collecting could provide tens of thousands of “lab-in-the-loop” simulations, just what an agent needs for scientific RL. Two months later, it’s matured into Minos, and today they’re releasing a full preprint https://github.com/Leash-Labs/minos-paper/blob/trunk/Minos_Leash.pdf, blog https://leashbio.substack.com/p/making-reality-replayable, and open dataset https://strelka.leash.bio/. We’ll be releasing a lightning podcast in a few days with an update. Stay tuned! We had a blast with Ian, probably one of our most fun episodes to date. For both complete chemistry novices and expert AI x Bio practitioners, we think you’ll get something out of this. Enjoy! We covered: * Why DELs traumatize everyone who touches them, and how Leash puts dozens of barcodes on each molecule to catch proteins that bind the DNA instead of the drug. * Ian’s rule that biology only scales two ways: taking pictures and sequencing DNA. * The Kaggle competition https://www.kaggle.com/competitions/leash-BELKA: about 4B measurements from a basement and about 2,000 teams, with nobody beating chance on an unseen chemical library. Also why Leash quietly feared co-folding would make them obsolete. * Clever Hans https://en.wikipedia.org/wiki/Clever_Hans in ChEMBL https://www.ebi.ac.uk/chembl/: models can guess a molecule’s chemist from its structure, and “who made it” predicts activity about as well as the structure does. * Hermes https://leashbio.substack.com/p/we-now-know-what-it-will-take, trained only on Leash’s own data, keeping pace with Boltz https://github.com/jwohlwend/boltz on public benchmarks. Plus the BRD4 molecule Boltz had already seen in the PDB. * Breaking the pocket https://leashbio.substack.com/p/taking-away-the-protein-cheat-sheet: engineering protein mutants at scale after catching Hermes memorizing “nicknames” for proteins instead of binding sites. * Lumacaftor https://en.wikipedia.org/wiki/Lumacaftor-like hits as an accidental detector for misfolded proteins. * The agent sandbox https://github.com/Leash-Labs/minos-paper/blob/trunk/Minos_Leash.pdf: LLMs that reason about SAR in plain English, hunting for the best binders among 6.5M molecules on a budget of 100. * The Recursion https://www.recursion.com/ mafia, and why its alumni companies all rhyme. * The Kosuri paradox https://x.com/srikosuri/status/2035866164325437894: if you can collect the data, why do you need the model? * Going from a DEL hit to good mouse PKPD in six months on an inflammation program, and the case for pills over antibodies on cytokine targets. * What stops a better-funded lab from outscaling them, and Ian’s by-fiat wish for massively parallel purification. This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.latent.space/subscribe https://www.latent.space/subscribe?utm_medium=podcast&utm_campaign=CTA_2

Transcript

0:00
1:49:56

23,884 words · 3 speakers

Topics

  • Protein
  • DNA
  • Dells
  • Hermes
  • Attention
  • Recursion
  • Mouse
  • Kaggle
  • XGBoost
  • SMILES

Page updated · transcript and summary are machine-generated