AutoZyme: An Autonomous Agentic Framework to Optimize Bioinformatics Software

Area of Science:

• Computational Biology
• Bioinformatics Software Engineering
• Scientific Computing

Background:

• Genomics and bioinformatics software face performance bottlenecks due to increasing biological dataset sizes.
• Manual optimization is expert-dependent and difficult to scale, hindering efficient data analysis.

Purpose of the Study:

• To introduce AutoZyme, an agentic framework for automated scientific software optimization.
• To address the scalability challenges in optimizing performance-critical bioinformatics tools.

Main Methods:

• AutoZyme identifies performance bottlenecks by building benchmarks and iteratively testing code changes.
• The framework retains optimizations that improve runtime while preserving functional output.

Main Results:

• AutoZyme improved runtime in over 95% of 45 evaluated functions without significant memory increases.
• For 38 functions from Seurat, Scanpy, and related packages, AutoZyme achieved a median runtime reduction of 8.52-fold, with some exceeding 676-fold.

Conclusions:

• AutoZyme offers a scalable solution for optimizing genomics and bioinformatics software.
• The framework and its optimized functions (AutoZyme-Library) enhance the efficiency of biological data analysis pipelines.