How do computational methods predict protein structure and dynamics?

Computational structure prediction has transformed with deep learning. AlphaFold2/ESMFold predict structures from sequence using attention mechanisms trained on PDB structures and evolutionary covariance. Accuracy approaches experimental resolution for many proteins. Molecular dynamics (MD) simulations model protein dynamics: force fields (AMBER, CHARMM, GROMACS) calculate atomic interactions; timescales reach milliseconds with specialized hardware (Anton) or enhanced sampling (metadynamics, replica exchange). Applications include: conformational sampling for drug design, binding free energy calculations, and mutational effect prediction. Limitations: difficulty with intrinsically disordered regions, conformational changes, complexes, and effects of PTMs or ligands. Integration with experimental data (cryo-EM densities, SAXS, NMR restraints) provides comprehensive structural understanding.