Recent Advances in Protein Structure Prediction (2019–2024): A Literature Review from Traditional Methods to AI Models

Abstract

Protein structure prediction (PSP) remains one of the most challenging and impactful problems in computa- tional biology. This review systematically examines the evolution of PSP methodologies, from traditional computational approaches to cutting-edge deep learning techniques. We begin with classical methods such as homology modeling and molecular dynamics, then explore machine learning-based approaches including neural networks and protein language models. Special emphasis is placed on revolutionary deep learning architectures like AlphaFold2 and RoseTTAFold, which have achieved remarkable accuracy in recent CASP competitions. We also discuss emerging directions in reinforcement learning for protein folding simulation and design. Throughout the review, we highlight key biological insights, computational innovations, and remaining challenges in the field.