Molence Bio 的 Mol-1 智能体可预测最高 60K 残基复杂蛋白组装,10 个评测组装中 9 个超过 Anthropic 结果。智能体将生物学知识带入建模,协调结构预测与组装并随结果调整策略,结构预测在单张 NVIDIA H20 上完成。8 个达到 TM-score ≥ 0.90,最大成功预测含 120 条蛋白链、TM-score 0.997。
2/15 🧩 Accurately predicting complex assemblies up to 60K residues, outperforming Anthropic 9 out of 10
Our long-running agent brings biological knowledge into the modeling process, coordinates structure prediction and assembly, and adapts its strategy as results come in, with structure prediction on a single NVIDIA H20.
Across 10 evaluated assemblies, the agent outperforms Anthropic’s reported results in 9 cases. Eight achieve a TM-score ≥ 0.90.
The largest successful prediction contains 120 protein chains and reaches a TM-score of 0.997.
#AIforScience
来源:Molence · x.com