Protein Folding: From DCA to AlphaFold

This lecture delves into the evolution of protein folding techniques, starting from DCA to the groundbreaking AlphaFold. The instructor explains the concept of pseudo-likelihood and its application in predicting amino acid sequences. The lecture explores the transition from physics-based potentials to co-evolutionary restraints, leading to the development of AlphaFold. It discusses the workflow of direct coupling analysis, the significance of multiple sequence alignment, and the role of deep learning in protein structure prediction. The instructor highlights the improvements brought by AlphaFold, its training process, and the utilization of attention mechanisms and transformers. The lecture concludes by emphasizing the importance of simple methods like DCA in exploring datasets and understanding the evolution of protein folding algorithms.