跳到正文
Molence· @MolenceBio · X·· 17 天前AI 评分44
AI 导读

12/15 🧩 分子胶预测必须同时正确处理蛋白质界面和分子胶姿态。 在 MGBench 上,Mol-1 的联合成功率为: • Seen:92.2% • High Homology:74.6% • Low Homology:28.6% 最强评估基线在 Low Homology 上达到 19.1%。 联合成功要求配体 RMSD < 2 Å、lDDT-PLI > 0.8 以及蛋白质界面 DockQ > 0.23。 #DrugDiscovery

正文 · 原文

12/15 🧩 Molecular-glue prediction has to get the protein interface and the glue pose right together.

On MGBench, Mol-1 achieves joint success rates of:
• Seen: 92.2%
• High Homology: 74.6%
• Low Homology: 28.6%

The strongest evaluated baseline reaches 19.1% in Low Homology.

Joint success requires ligand RMSD < 2 Å, lDDT-PLI > 0.8 and protein-interface DockQ > 0.23.

#DrugDiscovery

引用Molence@MolenceBio
11/15 💊 Stronger ligand prediction in unfamiliar binding environments. On Runs N’ Poses, Mol-1 leads the evaluated baselines across all eight similarity bins. In the lowest-similarity 0–20 bin, success is 36.1%, versus 24.6% for AlphaFold 3. Supplying binding-site information through Mol-1-constraint raises it to 63.9%. Success requires both ligand RMSD < 2 Å and lDDT-PLI > 0.8. #DrugDiscovery
在 X 查看被引用的帖子

来源:Molence · x.com