DrugHarness couples skill evolution with computational drug discovery agents
Original title: DrugHarness: An Evolvable Agent Harness for Computational Drug Discovery
Without evolution it beats reported MolClaw-CC affinity accuracy by 10.8 points; a compound hit 14.19 nM IC50.
DrugHarness is an agent harness that composes more than 80 scientific operations and induces skills from execution feedback. An independent review agent checks candidate skills before they guide later tasks. On MolBench, the unevolved system exceeds reported MolClaw-CC affinity-comparison accuracy by 10.8 percentage points. In a prospective HPK1 study, the authors report a synthesized compound with a cellular IC50 of 14.19 nM.
Unconfirmed: Potency, docking and benchmark gains are author-reported; independent replication is not stated.
- Evolution raises Qwen affinity-comparison accuracy from 62.2% to 81.1%.
- Optimization success rises from 79.5% to 87.2% on skill-development tasks.
- The compound is 2.64-fold more potent than the best input reference.
- The environment preserves links among candidates, results and artifacts.
- Products
- DrugHarness
- AI models
- Qwen
From the original article
1 min read at arxiv.orgComputational drug discovery requires agents to compose scientific operations and adapt to changing evidence and constraints. Here we introduce DrugHarness, an evolvable agent harness coupling composable scientific execution with conditional skill induction.
By Yuxing Chen · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.