ADMET-XSpec: A Tool for Systematic Cross-Species Data Integration in ADMET Prediction

The rapid expansion of in silico methodologies has reshaped modern drug discovery and toxicology research; however, robust prediction of ADMET endpoints remains limited by the scarcity and heterogeneity of experimental data. In particular, toxicological data sets are often fragmented across species, complicating the development of reliable and generalizable machine learning models. To address this challenge, we introduce a dedicated Python-based computational package, ADMET-XSpec, designed for the systematic development, training, and evaluation of ML models with an explicit consideration of interspecies data integration. The framework enables controlled incorporation of chemical space originating from different species and different assay types, allowing users to flexibly construct single-species models, as well as models augmented with cross-species information. This design facilitates systematic investigation of how additional data from other organisms influence model performance without imposing assumptions inherent to specific transfer learning paradigms. By supporting standardized preprocessing, scalable integration of heterogeneous data sets, and rigorous benchmarking, the proposed tool provides a unified environment for studying cross-species effects in ADMET modeling. Overall, this work delivers a practical resource for the ADMET modeling community and offers insights into how interspecies and interassay data integration can improve model robustness and generalizability while clarifying the conditions under which cross-species and cross-assay data information is beneficial for predictive toxicology. The package is freely available at https://github.com/hubertrybka/admet-xspec. ADMET-XSpec advances the state of the art by providing the first dedicated framework for controlled interspecies and interassay data integration in ADMET modeling, offering quantitative guidance on when and how cross-species and cross-assay data improve predictive performance.