BioTester: an AI-driven automated testing framework for identifying potential quality risks in bioinformatics software.

Bioinformatics software plays a critical role in clinical applications such as cancer screening and genetic disease diagnosis, where comprehensive quality management is essential for ensuring the accuracy and reliability of downstream analysis. However, current validation practices rely heavily on manually designed simulation experiments, which are labor-intensive and limited in their ability to systematically identify potential quality risks under certain scenarios. In this study, we first construct a benchmark by simulating subtle implementation-level defects in bioinformatics programs. We then propose BioTester, an oracle-based automated testing framework that integrates software testing techniques to support more comprehensive quality assessment of bioinformatics software. BioTester integrates retrieval-augmented LLMs with a differential testing strategy to address the long-standing oracle problem in bioinformatics software testing, demonstrating superior defect-detection performance over existing methods on the constructed benchmark. Finally, applying BioTester to real-world bioinformatics software demonstrates its practical effectiveness and highlights the value of automated testing as a generalizable complement to existing validation practices for improving software reliability in biomedical applications.