HSADab2025: AI-powered Modelling of Human Serum Albumin - A significant update to the HSADab database, incorporating AI-powered modelling to enhance our understanding of Human Serum Albumin (HSA) interactions. The new webserver hsadab.cn offers instant prediction of HSA binding affinities for drug-like molecules, a comprehensive affinity and structure database, and deep-learning assisted docking structures. - The database now includes the most up-to-date thermodynamic data (up to June 2024) and uses advanced machine learning predictors, such as fine-tuned large language models and graph-based neural networks, to achieve chemical accuracy in binding affinity predictions using just the SMILES representation of molecules. - A detailed analysis of HSA's protein conformational space reveals a clear distinction between bound and unbound states, providing crucial insights for docking studies. The research also benchmarks traditional docking protocols on HSA, identifying PLANTS as a top performer, which contrasts with the widespread use of AutoDock derivatives. - The docking bank has been significantly expanded with global docking results from state-of-the-art deep-learning tools like DiffDock, AF3, Boltz-1, and Chai-1, offering a more comprehensive structural basis for HSA-ligand interactions and enabling a deeper understanding of the binding process. - The study demonstrates the superiority of ML predictors over traditional physics-based screening methods in estimating HSA binding strength, with the ML models achieving near chemical accuracy without requiring 3D interaction descriptions of protein-ligand complexes. - All data entries, including the updated affinity, docking, and structure databanks, as well as the SMILES-to-affinity ML predictors, are freely available online, facilitating further research and applications in the field of HSA-drug interactions. 💻Code: github.com/proszxppp/HSADab 📜Paper: doi.org/10.26434/chemrxiv-20… #HumanSerumAlbumin #AIPoweredModelling #DrugDiscovery #MachineLearning #ProteinLigandInteractions #OpenSourceDatabase