Intology 推出 Locus：自动化分子设计的 AI 研发系统
Intology 做了个自动跑科学实验的系统 Locus，在分子设计任务上比 Claude Code 和 Codex 还强，制药圈的可以看看。
Intology 公司发布自动化科研系统 Locus，用于分子设计相关的 AI 模型研发。系统可自主提出实验方案并并行运行，覆盖分子性质预测、用 Boltz-2 微调小分子药效预测、以及基于 BoltzGen 和 ESMFold2 的抗体设计工作流。在保留测试集上，Locus 的表现超过 Claude Code 和 Codex 等通用智能体。它能在有限数据下自主把 Boltz-2 这类基于公开数据训练的模型适配到私有靶点上，替代以往需要专业 ML 团队完成的工作。
Highly recommended read. Very few companies I know have achieved this leap in scaling autonomous AI R&D in science. Intology claims their Locus system sets the frontier for automating molecular design. That covers molecular property prediction, potency prediction with Boltz-2, and antibody design workflows. What’s the challenge with automating AI R&D? With so little data and noisy measurements, it's easy for an agent to overfit and fail on new molecules. On held-out test data, Locus held up better than Claude Code and Codex. Under the hood, Locus proposes experiment ideas and runs several of them in parallel on purpose-built infrastructure. It picks the next round of experiments based on evaluation feedback. Proprietary data is where this gets really interesting. Models like Boltz-2 are trained mostly on public data. Real drug discovery runs on private targets and chemical series that can look very different from that training data. Locus adapts these models to new targets with limited data on its own, work that usually needs expensive ML expertise. Intology @intology Locus sets the frontier for automating molecular design Our automated research system Locus develops scientific AI models that outperform SOTA expert solutions Unlocking the potential of scientific AI requires leveraging proprietary data. Locus achieves this autonomously 🧵👇 -- We test Locus across a number of tasks related to molecular scoring and design: training molecular property prediction models for a proprietary task, fine-tuning Boltz-2 to predict small-molecule potency, and building antibody design workflows around BoltzGen, Boltz-2, ESMC, and ESMFold2. Each of these settings are real-world AI for science use cases, where tasks are often out-of-distribution for existing foundation models and limited in data, while featuring expensive and noisy evaluators. Compared to state-of-the-art baselines and LLM agents set to optimize these tasks, Locus achieves higher performance. Additionally, Locus generalizes better despite limited data, finds more sample efficient solutions, and explores a higher diversity of solutions. 🔗 View Quoted Tweet 💬 4 🔄 1 ❤️ 19 👀 2261 📊 5 ⚡
