Protein structure: Has Levinthal's paradox “folded”?

An annual competition among computational biochemists is held by an organization called Critical Assessment of Protein Structure Prediction (CASP). The winning 2020 entrant was from DeepMind, a British machine learning-based company. This team tweaked their earlier work into refinement that outpaced not only the competing platforms but also previous efforts in this mode protein folding field. notion linear arrays amino acids function because polymer's folded structure arose early field.1, 2 Key discoveries were made Alfred Mirsky (Figure 1) and later Christian Anfinsen 2). These pioneering experiments demonstrated acid side chain interactions dictated final structure, epistemologically erasing era which zephyrs vitalism still at play, proteins, whether colloids, glob(ule)s, or whatever shrouded mystery. Once biochemistry biosynthesis had been established 1950s, few astute recognized nascent polypeptide might start to fold before it finished, while others (the majority) focused on how product would fold. There empirical work, some hapless much instructive, yet seemed clear problem's combinatorial dimension great challenge; perhaps better adjective “astonishing.” Another term used time “unsolvable.” Sometimes takes oddball move things science, I contend molecular biologist Cyrus Levinthal one 3) he did that. In 1969, published typical could have something like 10300 possible conformations, quipped solving them all take longer than age universe.3 his subsequent career, put formidable scientific energy applying computer science biology, long terms “computational biology” “structural come our vernacular. Just as pathways through navigates potential folding, so too field itself discrete steps. One most important finding process involves “incorrect” forays, defined kinetic tentativeness difference state.4 senior author paper, Charles Tanford 4), has favorite mine expository lucidity papers. His article hydrophobic effect pedagogical masterpiece.5 DeepMind system, AlphaFold, “cheats” bit starts with empirically determined structures training templates, then roars descendent perturbations, “educating” “improving” every step. It thus does find theoretical different respect machine-learning based game strategies.6 imagine next step, interacting pairs proteins probed, since vast majority cell's operate heterotypic complexes.7 Meanwhile, these latest advances usher new enormous for drug design, enzyme engineering, insights evolution.