Protein–Protein Interface Topology as a Predictor of Secondary Structure and Molecular Function Using Convolutional Deep Learning

Abstract

To power the specific recognition and binding of protein partners into functional complexes, a wealth of information about the structure and function of the partners is necessarily encoded into the global shape of protein-protein interfaces and their local topological features. To identify whether this is the case, this study uses convolutional deep learning methods (typically leveraged for 2D image recognition) on 3D voxel representations of protein-protein interfaces colored by burial depth. A novel twostage network, fed with voxelizations of each interface at two distinct resolutions, achieves balance between performance and computational cost. From the shape of the interfaces, the network tries to predict the presence of secondary structure motifs at the interface and the molecular function of the corresponding complex. Secondary structure and certain classes of function are found to be very well predicted, validating the hypothesis of interface shape as a conveyor of higher-level information. Interface patterns triggering the recognition of specific classes are also identified and described.

Using machine‐learning‐driven approaches to boost hot‐spot's knowledge

“…These are extremely useful for ML applications which depend on the selection of key features that encode the main characteristics of the biological problem at hands. 197 Usually, researchers will provide a high number of local and global characteristics/features and let the algorithm choose by itself the ones that provide the higher discriminatory power. This learning depends not only on feature correlation but also on their encoding, renormalization and mix between different formats.…”

“…The increased number of features that can be considered brings however challenges in defining their relative weight. 197 The used features in the development of different data-driven models typically fall into two broad categories:…”

“…This learning depends not only on feature correlation but also on their encoding, renormalization and mix between different formats. The increased number of features that can be considered brings however challenges in defining their relative weight 197 …”

“…The current big data era is resulting in a huge number of rich sources of molecular level information for proteins. These are extremely useful for ML applications which depend on the selection of key features that encode the main characteristics of the biological problem at hands 197 . Usually, researchers will provide a high number of local and global characteristics/features and let the algorithm choose by itself the ones that provide the higher discriminatory power.…”