Protein domain-domain interaction prediction via deep neural networks
Proteins are one of the building blocks of life, and their structures and functions maintain most of the cellular processes. Determination of protein structures is not only critical for understanding its working mechanism but also vital for protein engineering and drug design. Although many experimental approaches exist to reveal the structures of proteins, limitations of experimental methods led researches to develop computational approaches to determine structures. Implementation of machine learning algorithms provided great improvements in the protein structure prediction area for small and medium-sized proteins. On the other hand, for large proteins, determination of the structure of the overall protein complex remains a big challenge. One common approach for determination of large protein complexes is to determine the structures of protein subunits (i.e. domains) individually and arrange their positions and orientations correctly. For this purpose, we used convolutional neural networks to predict the distance potentials between the monomers of the target domain pairs. Successful distance potential predictions allowed us to generate correct interfaces between the domain pairs. This method will help to determine the structures of large, multi-domain protein complexes that can result in understanding their function better and lead to design successful experiments for protein engineering.
Healthy world