Proteins perform a vast number of functions in cells including signal transduction, DNA replication, catalyzing reactions, etc. Engineering and designing proteins for specific structure and function not only deepen our understanding of the protein sequence-structure relationship, but also have wide applications in chemistry, biology and medicine. Over the past three decades, remarkable successes have been achieved in protein design, in which some of the designs were guided by computational methods. Examples of some recent successful computational protein designs include novel folds, enzymes, vaccines and antibodies. Deep-learning neural network, as a machine learning technique, is becoming increasingly powerful with the development of new algorithms and computer hardware, and has been applied to learning massive data sets in a variety of fields such as image recognition or language processing. The advantage of using deep neural network is that it can learn high-order features from simple input data such as atom coordinates and types. The technical details such as network architecture, data representations vary from application to application, but the fundamental requirement of applying deep neural network is the availability of a large amount of data. With the aforementioned rich protein structure data available, it is promising to apply deep neural network in computational protein design. Here you have a recent example published by Huddy et al in Nature (doi.org/10.1038/s41586-024-07188-4), showing the cryoEM structure of a computationally-designed T3 tetrahedral protein nanocage (PDB code: 8TL7)

#molecularart ... #nanocage ... #design ... #ai ... #tetrahedral

Structure rendered with 3D Protein Imaging, post-processed with Stylar and depicted with @corelphotopaint