Lab-in-the-Loop at Compute Speed — Ian Quigley, Leash Bio
https://www.latent.space/p/lab-in-the-loop-at-compute-speed📌 DNA 編碼庫養出的乾淨資料，正在教 AI 做藥物化學推理
TL;DR：Leash Bio 用高通量 DNA Encoded Library 資料訓練模型，並讓 LLM agent 在數百萬分子中做可驗證的科學推理。
多數人一聽到 DNA Encoded Library（DEL）就搖頭，因為這種高通量蛋白質與小分子交互作用資料出了名的髒。但 Leash Bio 找到辦法把它洗乾淨到可以拿去做大規模建模，而他們最新的動作，是把這批資料變成 AI agent 的科學推理訓練場。
🤔 DEL 為什麼讓人頭痛
DEL 是一種能大量收集蛋白質與小分子交互作用數據的方法，但常見的問題是假訊號——很多時候抓到的其實是蛋白質結合到 DNA 本身，而不是結合到要測試的藥物分子。Leash 的作法是在每個分子上貼上幾十個 barcode，用來辨識並排除這類偽陽性結果。
🧩 一場沒人贏過隨機猜測的 Kaggle 競賽
Leash 辦過一場 Kaggle 競賽，用大約 40 億筆量測資料、吸引約 2,000 支隊伍參賽，結果是沒有任何隊伍在一個從未見過的化學庫上打贏隨機猜測——直接說明 protein-ligand binding 預測在業界仍未真正解決。素材也提到 Leash 原本暗自擔心 co-folding 技術會讓他們的方法過時。
🧩 「Clever Hans in ChEMBL」：模型學到的是作者，不是化學
團隊發現一件挺諷刺的事：模型可以單靠分子結構猜出這是哪位化學家合成的，而「這個分子是誰做的」這個資訊，預測活性的準確度幾乎跟分子結構本身一樣好。這等於揭露了像 ChEMBL 這類公開資料庫裡隱藏的資料洩漏問題——模型很可能不是在學化學，而是在學「認人」。
🧩 Hermes 與 Breaking the Pocket
Leash 用自家資料訓練出 Hermes 模型，在公開 benchmark 上追得上 Boltz 的表現，但素材也點出 Boltz 比較的 BRD4 分子其實早就出現在 PDB 資料庫裡，等於已經「看過答案」。更進一步，團隊發現 Hermes 其實是記住了蛋白質的「暱稱」，而不是真的學會 binding pocket，於是工程化大量蛋白質突變體來破解這個捷徑。另一個意外收穫是，像 Lumacaftor 的命中分子反而變成偵測蛋白質是否摺疊錯誤的指標。
🧩 把分子推理場交給 LLM agent
團隊最近打造了一個 agent sandbox，讓 LLM 用「plain English」討論 SAR（structure-activity relationship），在 650 萬個分子裡、只有 100 次嘗試的預算下尋找最好的 binder。這套 lab-in-the-loop 模擬環境後來發展成 Minos，團隊同步發布了 preprint、部落格文章與開放資料集。
💡 Kosuri Paradox：資料都收集到了，為什麼還需要模型？
素材提到的 Kosuri paradox 點出一個張力：當實驗本身已經能產出足夠大量的數據，模型存在的價值到底是什麼？從 Leash 的脈絡看，答案似乎是模型要把這些高通量數據轉成可泛化、可推理的知識，而不只是把資料庫當成查表工具。
📊 從 DEL 命中到真實藥物開發的時間線
素材提到在一個發炎（inflammation）相關計畫中，團隊從 DEL 命中分子到在小鼠身上取得不錯的 PKPD（藥物動力學／藥效學）結果，只花了六個月；文中也提到在 cytokine 靶點上，小分子藥丸相對於抗體藥物可能更具優勢的論點。
⚠️ 護城河與未解問題
素材並未給出 Minos 模型的具體準確率或完整 benchmark 數字，對於「什麼能阻止資源更雄厚的實驗室靠規模超越 Leash」這個問題，也只提到 Ian Quigley 自己對「大規模平行純化」有一個尚未實現的期望，而非具體解法。
🎯 實務啟示
對做 AI x Bio 的工程師，這個案例給出一個清楚的心法：把你的問題對應成一個 combinatorial library 加上 next-generation sequencing 的 readout 方式；同時在訓練任何活性預測模型前，務必檢查資料裡是否藏著像「作者是誰」這種看似無關、實際上會被模型拿來走捷徑的隱藏特徵。
🔗 來源
- 標題：Lab-in-the-Loop at Compute Speed — Ian Quigley, Leash Bio
- 作者／機構：Latent Space（訪談對象：Ian Quigley, Leash Bio）
- 連結：https://www.latent.space/p/lab-in-the-loop-at-compute-speed
#LeashBio #DrugDiscovery #DNAEncodedLibrary #AIforScience #ProteinLigandBinding #ScientificRL #LLMAgent #ComputationalChemistry #Minos #Bioinformatics
原始資料 Latent Space · 收集於 2026-10-14
摘要原文
A few years ago I stumbled upon Leash Bio’s Substack , and was hooked. Be it their Kaggle competition that showed that no, we really haven’t solved protein-ligand binding , or their crazy discovery that knowing a molecule’s author is perhaps all you need to know to predict how good it is . Their substack — or at least Owl Posting ’s 1 deep dive 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 from Leash Bio, and pick his brain. Leash specializes in DNA Encoded Libraries (aka “DELs”), a way of collecting ridiculously high-throughput interaction data between proteins and small molecules. 2 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 , blog , and open dataset . 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. 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 : 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 in 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 , trained only on Leash’s own data, keeping pace with Boltz on public benchmarks. Plus the BRD4 molecule Boltz had already seen in the PDB. Breaking the pocket : engineering protein mutants at scale after catching Hermes memorizing “nicknames” for proteins instead of binding sites. Lumacaftor -like hits as an accidental detector for misfolded proteins. The agent sandbox : LLMs that reason about SAR in plain English, hunting for the best binders among 6.5M molecules on a budget of 100. The Recursion mafia, and why its alumni companies all rhyme. The Kosuri paradox : 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. The hands-down best bio blogger on the internet. As always, the secret here is to map your problem to a combinatorial library and a readout using Next Generation Sequencing. We talk about this in depth in the podcast!
由 claude-code/sonnet 自動生成