Protein Structure Prediction Methods.

Protein structure prediction, a fundamental challenge emerging from the protein folding problem, forms the basis of modern computational biology. This field addresses the critical question of how the amino acids sequence determines its three-dimensional structure, a relationship critical to understanding biological function. Over the last four decades, methodologies have evolved from template-based modeling (TBM) and free modeling (FM) to advanced hybrid and end-to-end deep learning approaches. TBM explores sequence homology and threading to predict structures based on a known template, while FM applies physics-based principles to navigate the rugged energy landscape that governs protein folding, predicting de novo stable native conformations. Recent breakthrough methods in protein structure prediction include hybrid methods that integrate physics, bioinformatics, and machine learning, as well as end-to-end methods such as AlphaFold2 and RoseTTAFold, which have revolutionized the field by using neural networks to directly predict atomic coordinates from sequences, achieving near-experimental accuracy. Protein language models further advance the field by learning sequence-structure-function relationships directly from amino acid sequences, bypassing the need for multiple-sequence alignments. These innovations address the sequence-structure-function paradigm and find applications in drug discovery, enzyme engineering, and disease research. This chapter explores the principles, advances, and transformative impact of these methodologies on the structural biology field.