Methods Mol Biol. 2026;3061:333-345. doi: 10.1007/978-1-0716-5420-0_20.
ABSTRACT
Recent advances in deep-learning-based protein structure prediction have transformed access to three-dimensional models of biological targets directly from sequence information. Methods such as AlphaFold and emerging co-folding approaches have greatly expanded structural coverage of the proteome, enabling structure-based drug design (SBDD) for targets lacking experimentally determined structures. Integration of AI-derived protein structures with quantum-mechanical methods such as the Fragment Molecular Orbital (FMO) approach provides a powerful and complementary strategy, in which AI-based models deliver rapid structural information, while FMO leverages these structures to enable quantitative, physics-based decomposition of protein-ligand and protein-protein interactions. FMO yields interaction strengths in kcal/mol, characterizes the chemical nature of interactions (electrostatic or hydrophobic), and enables subsystem analysis to evaluate the contributions of different energetic terms, including polarization and desolvation, to binding. Together, these approaches transform predicted structures into energetically interpretable models that directly support structure-activity relationship (SAR) analysis, binding-mode validation, and prospective structure-based drug design.
PMID:42604913 | DOI:10.1007/978-1-0716-5420-0_20