Process-Reward Tactic Evolution for Long-Horizon Bioinformatics Workflows

2026-06-18

AI Summary

This work addresses the limitations of large language model (LLM) agents in executing long-horizon bioinformatics workflows—specifically, their lack of reliable interaction, process validation, and reusable strategic planning. To overcome these challenges, the authors propose a training framework built upon the Galaxy platform that incorporates a process-supervision-driven strategy evolution mechanism. The approach decomposes complex workflows into learnable, reusable tactical units and integrates curriculum learning, a process validator, trajectory distillation, and tactical library retrieval to construct an Agent Gym training environment alongside the BioWorkflow Bench evaluation benchmark. Experimental results demonstrate that the proposed method significantly outperforms existing baselines in workflow completion rate, biological correctness, and execution efficiency, achieving, for the first time, efficient and reliable automation of bioinformatics workflows supported by a reusable strategy library.

Abstract

LLM agents can write code and call tools, but reliable bioinformatics work requires long-horizon interaction with workflow software, typed data objects, provenance, and biological checks. We study this setting through Galaxy workflow execution. The agent must explore task data, construct or adapt an executable workflow DAG, bind inputs and dataset collections, monitor execution, debug failures, and validate biological outputs. We propose Process-Reward Tactic Evolution, a Galaxy-based training framework that turns verified workflow rollouts into reusable \tactics. During training, agents practice on curriculum-organized Galaxy tasks in Agent Gym; process verifiers score workflow construction, software interaction, execution, and biological correctness; successful and failed traces are distilled into a tactic library. At inference, the trained executor, Process-Reward Tactic Evolution, uses this library to execute held-out peer reviewed Galaxy workflow converted BioWorkflow Bench and BioAgent Bench tasks in isolated environments. The paper evaluates whether process-supervised tactic accumulation improves long-horizon bioinformatics workflow completion, biological correctness, and execution efficiency over no-memory and reflection-style baselines.

Problem

• long-horizon workflows
• bioinformatics
• workflow execution
• biological correctness
• LLM agents

Innovation

• Process-Reward Tactic Evolution
• long-horizon workflow
• tactic library
• process supervision
• Galaxy platform