Molence· @MolenceBio · X·· 17 天前AI 评分51
AI 导读
Molence 称 Mol-1 驱动的长时智能体在单张 NVIDIA H20 上完成 60,120 原子分子组装的结构预测,10 项评测组装中有 9 项优于 Anthropic 报告的结果。其中 8 项达到 TM-score ≥ 0.90。同一线程的前一条称 Mol-1 在 FoldBench-AB 上取得 75.0% 中位成功率(DockQ > 0.23),并给出 GitHub 仓库与试用入口。
正文 · 原文
2/15 🧩 A 60,120- molecular assembly, with structure prediction on a single NVIDIA H20.
Mol-1-based long running agent outperforms Anthropic’s reported results on 9 of 10 evaluated assemblies.
Eight achieve TM-score ≥ 0.90. #AIforScience
1/15 🧬 Mol-1 on FoldBench-AB: 75.0% median success at DockQ > 0.23, matching IsoDDE’s published 75.58%. We evaluated all 201,376 five-seed combinations from 32 seeds and 62.22%matched or exceeded 75% success, showing robust performance. Explore Mol-1: https://github.com/molence/Mol-1 Try it: https://lab.molence.com #AIforScience #DrugDiscovery在 X 查看被引用的帖子
来源:Molence · x.com