Badapple 2.0: An Empirical Predictor of Compound Promiscuity, Updated, Modernized, and Enhanced for Explainability

Modern drug discovery and chemical biology research relies heavily on analyzing bioassay data. One of the many challenges in bioassay data analysis is identifying false trails, i.e., chemical compounds which initially appear to have desirable activity but are found to be problematic upon further investigation. Badapple (BioAssay-Data Associative Promiscuity Pattern Learning Engine) was initially released in 2012, to help researchers identify promiscuous compounds and thus avoid a common source of these false trails, and since then has been well regarded and used by the global scientific community. Through an effort involving software engineering, cheminformatics, and biomedical data science we have developed Badapple 2.0, a major update with a complete code-rewrite, updated and expanded assay data sets, enhanced functionality, scalability, and metadata, supporting improved explainability and richer bioactivity analyses. Badapple 2.0 was motivated and funded by an ongoing AI/ML-empowered antialphaviral discovery program, but has broad applicability for improving the efficiency of many other early stage drug discovery campaigns.