2021/05/16 by Élodie Laine, Laine, Elodie, Stephan Eismann +5 · 1 citation
Biochemistry, Genetics and Molecular Biology · #Biomolecules (q-bio.BM) #FOS: Biological sciences #FOS: Computer and information sciences #Genomics and Phylogenetic Studies #Machine Learning (cs.LG) #Machine Learning in Bioinformatics #Protein Structure and Dynamics
paper · pdf · doi:10.48550/arxiv.2105.07407
openalex publication_date 2021/05/16 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
The potential of deep learning has been recognized in the protein structure\nprediction community for some time, and became indisputable after CASP13. In\nCASP14, deep learning has boosted the field to unanticipated levels reaching\nnear-experimental accuracy. This success comes from advances transferred from\nother machine learning areas, as well as methods specifically designed to deal\nwith protein sequences and structures, and their abstractions. Novel emerging\napproaches include (i) geometric learning, i.e. learning on representations\nsuch as graphs, 3D Voronoi tessellations, and point clouds; (ii) pre-trained\nprotein language models leveraging attention; (iii) equivariant architectures\npreserving the symmetry of 3D space; (iv) use of large meta-genome databases;\n(v) combinations of protein representations; (vi) and finally truly end-to-end\narchitectures, i.e. differentiable models starting from a sequence and\nreturning a 3D structure. Here, we provide an overview and our opinion of the\nnovel deep learning approaches developed in the last two years and widely used\nin CASP14.\n