With the rapid advancement of Artificial Intelligence(AI)technology,cutting-edge protein structure prediction methods have emerged,exemplified by AlphaFold2 and RoseTTAFold[1].These methods have revolutionized our und...With the rapid advancement of Artificial Intelligence(AI)technology,cutting-edge protein structure prediction methods have emerged,exemplified by AlphaFold2 and RoseTTAFold[1].These methods have revolutionized our understanding and utilization of protein structures in biological research.While their primary focus lies in predicting the 3D structures of separate protein molecules,the potential of native proteins to perform essential functions through the formation of various"biomolecular assemblies"is overlooked.These assemblies involve interactions with other biomolecules including nucleic acids,polysaccharides,metals,small ligands,etc.The complex bonding mechanisms—ranging from covalent and noncovalent bonding to metal chelation,etc.—between different molecular units play critical roles in maintaining biological activities of these assemblies.Thus,effectively predicting binding interactions between proteins and other(bio)molecules within one biomolecular assembly remains an outstanding challenge.Recently,one groundbreaking work published in Science by Baker et al.released two upgraded deep learning tools:RoseTTAFold All Atom(RFAA)and RFdiffusion All-Atom(RFdiffusionAA)[2],which significantly broaden the in silico construction scope for biomolecule assemblies,enabling researchers to explore complex interactions beyond individual protein structures(Fig.1).展开更多
文摘With the rapid advancement of Artificial Intelligence(AI)technology,cutting-edge protein structure prediction methods have emerged,exemplified by AlphaFold2 and RoseTTAFold[1].These methods have revolutionized our understanding and utilization of protein structures in biological research.While their primary focus lies in predicting the 3D structures of separate protein molecules,the potential of native proteins to perform essential functions through the formation of various"biomolecular assemblies"is overlooked.These assemblies involve interactions with other biomolecules including nucleic acids,polysaccharides,metals,small ligands,etc.The complex bonding mechanisms—ranging from covalent and noncovalent bonding to metal chelation,etc.—between different molecular units play critical roles in maintaining biological activities of these assemblies.Thus,effectively predicting binding interactions between proteins and other(bio)molecules within one biomolecular assembly remains an outstanding challenge.Recently,one groundbreaking work published in Science by Baker et al.released two upgraded deep learning tools:RoseTTAFold All Atom(RFAA)and RFdiffusion All-Atom(RFdiffusionAA)[2],which significantly broaden the in silico construction scope for biomolecule assemblies,enabling researchers to explore complex interactions beyond individual protein structures(Fig.1).
