MolClaw: An Autonomous Agent with Hierarchical Skills for Drug Molecule Evaluation, Screening, and Optimization

2026-04-02

bioRxiv

AI Summary

Current AI agents struggle to coordinate multi-step, multi-tool workflows in complex drug molecule screening and optimization, limiting their performance. This work proposes the first three-tiered hierarchical skill architecture tailored for drug discovery—comprising tool, workflow, and domain layers—that integrates over 30 computational chemistry and AI tools and embeds domain-knowledge-driven mechanisms for planning, validation, and reflection. Building upon this framework, the authors introduce MolBench, a comprehensive benchmark featuring tasks requiring 8 to 50+ tool invocations. Experiments demonstrate that the proposed approach achieves state-of-the-art performance across MolBench, and ablation studies confirm that its superiority stems from its structured workflow coordination capability, revealing such coordination as a critical bottleneck in AI-driven drug discovery.

Abstract

Computational drug discovery, particularly the complex workflows of drug molecule screening and optimization, requires orchestrating dozens of specialized tools in multi-step workflows, yet current AI agents struggle to maintain robust performance and consistently underperform in these high-complexity scenarios. Here we present MolClaw, an autonomous agent that leads drug molecule evaluation, screening, and optimization. It unifies over 30 specialized domain resources through a three-tier hierarchical skill architecture (70 skills in total) that facilitates agent long-term interaction at runtime: tool-level skills standardize atomic operations, workflow-level skills compose them into validated pipelines with quality check and reflection, and a discipline-level skill supplies scientific principles governing planning and verification across all scenarios in the field. Additionally, we introduce MolBench, a benchmark comprising molecular screening, optimization, and end-to-end discovery challenges spanning 8 to 50+ sequential tool calls. MolClaw achieves state-of-the-art performance across all metrics, and ablation studies confirm that gains concentrate on tasks that demand structured workflows while vanishing on those solvable with ad hoc scripting, establishing workflow orchestration competence as the primary capability bottleneck for AI-driven drug discovery.

Problem

• computational drug discovery
• molecule screening
• molecule optimization
• AI agent
• workflow orchestration

Innovation

• hierarchical skill architecture
• workflow orchestration
• autonomous agent
• computational drug discovery
• MolBench