Abstracts
Hybrid AI and Open Source for Drug Design
Andrea Volkamer
Data Driven Drug Design, Saarland University, Saarbrücken [DE]
Addressing the complexities of drug design – exemplified here on human protein kinases – necessitates innovative approaches that blend artificial intelligence (AI) with domain expertise. With over 6,000 human kinase structures available in the PDB and around 70 small molecule kinase inhibitors, persistent challenges such as drug promiscuity, resistance, and unexplored kinase territories remain.
This presentation showcases open-source approaches that integrate domain knowledge into AI frameworks to overcome data scarcity issues and enhance model generalization. Leveraging openly available kinase data, we demonstrate how hybrid AI and classical methods can generate new insights and foster community engagement. The TeachOpenCADD platform [1,2] serves as a versatile tool for orchestrating diverse computer-aided drug design (CADD) tasks, exemplified on individual kinases. We introduce freely available resources to support kinase research and incorporate structural data to enhance prediction accuracy and guide drug discovery efforts. These resources include active learning on the KinFragLib kinase fragment library for inhibitor design [3], and structure-based deep learning approaches for affinity prediction [4].
These projects highlight how hybrid AI approaches fuse experimental with computational insights to advance the frontier of kinase-focused drug discovery.
References:
[1] S. Dominique, et al., TeachOpenCADD 2022: Open Source and FAIR Python Pipelines to Assist in Structural Bioinformatics and Cheminformatics Research. Nucleic Acids Research, 2022. https://doi.org/10.1093/nar/gkac267
[2] M. Backenköhler, et al., TeachOpenCADD goes Deep Learning: Open-source Teaching Platform Exploring Molecular DL Applications. ChemRxiv, 2023. https://doi.org/10.26434/chemrxiv-2023-kz1pb
[3] S. Dominique, et al., KinFragLib: Exploring the Kinase Inhibitor Space Using Subpocket-Focused Fragmentation and Recombination. Journal of Chemical Information and Modeling, 2020. https://doi.org/10.1021/acs.jcim.0c00839
[4] M. Backenköhler, J. Groß, V. Wolf, A. Volkamer, Guided docking as a data generation approach facilitates structure-based machine learning on kinases. Journal of Chemical Information and Modeling, 2024, 64, 10, 4009–4020 https://pubs.acs.org/doi/10.1021/acs.jcim.4c00055