RANQSAR: a standalone open-source application for reproducible machine learning-based QSAR analysis

Despite the importance of quantitative structure-activity relationship (QSAR) modeling in ligand-based drug discovery, the use of reproducible and standardized workflows for QSAR model development often involves multiple software tools and programming. Here, we introduce RANQSAR, an open-source standalone desktop application providing a reproducible environment for machine learning-based regression QSAR modeling. RANQSAR integrates molecular descriptor calculation, dataset split, model construction, validation, assessment of applicability domains, estimation of uncertainties, benchmarking, and prospective predictions within one user-friendly graphical interface. Several molecular representations based on Morgan fingerprints, RDKit-generated physicochemical descriptors, and combinations of both are available in RANQSAR along with different algorithms suitable for regression. Deterministic execution of the workflow in RANQSAR by means of seeding and automatic validation, including cross-validation, external validation, y-randomization test, and applicability domain analysis, ensures proper QSAR practices. We showcase the usefulness of RANQSAR using acetylcholinesterase inhibitors as a model study case and demonstrate that the platform allows one to develop a reproducible regression QSAR model and carry out a comparative benchmarking. We envisage RANQSAR as a practical application for academics, QSAR students, and computational drug discovery experts to use when working on QSAR models. RANQSAR is distributed as open-source software and provided together with example datasets and documentation.