The 3D atomic structure of a protein provides key insights into understanding its function. For this reason, tremendous research efforts have been invested to decipher the structural code of protein sequences. Unfortunately, to deduce a protein’s structure from its amino acid sequence, i.e., the protein structure prediction problem, turns out very challenging without an easy back-of-the-envelope answer. From a theoretical and computational prospective, the quest for solving the protein structure prediction problem has driven many breakthroughs in the field of computational biology. For practical applications, comparative modeling based on a related structural template is the most successful approach. However, such an approach is not applicable when no approximate template is available or, more likely, is too subtle too be detected. We introduce a new approach, DESTINI that combines deep-learning with comparative, template-based, structural modeling for proteins. Through large-scale benchmarking tests on over one thousand different proteins, we demonstrate that DESTINI breaks the “glass-ceiling” that limited the success of previous knowledge-based approaches. The key improvement by DESTINI lies in its more accurate protein residue/residue predictions through the recognition of contact patterns, which in turn drives structural assembly towards the correct protein fold. Such an improvement is another step towards a general solution of the protein structure prediction problem.
Notice: This server is freely available to all academic and non-commercial users.
Commercial users – to use this server, or request an evaluation copy, please send an email to Dr. Jeffrey Skolnick: skolnick@gatech.edu.
If you find this service useful, please cite the following paper: Mu Gao, Hongyi Zhou, Jeffrey Skolnick. 2019. DESTINI: A deep-learning approach to contact-driven protein structure prediction. Scientific Reports. 9: 3514. https://doi.org/10.1038/s41598-019-40314-1. PDF
Submit your protein sequence to DESTINI
Download benchmark data sets and results.
Please send questions and comments to Dr. Mu Gao (mu.gao@gatech.edu).