EN FR
EN FR


Section: New Results

Analysis of a deep-learning architecture for fold recognition

Deep learning has recently demonstrated outstanding capabilities in classical pattern recognition problems. It has also obtained a tremendous success in the very recent protein structure prediction tasks. This work studies recurrent structural patterns in protein structures recognized by a deep neural network. We demonstrated that neural networks can automatically learn a vast amount of chemo-structural features with only a very little amount of human supervision. For example, our architecture correctly learns atomic, amino acid, and also higher-level molecular descriptors. The network architecture and the results are available at https://team.inria.fr/nano-d/software/Ornate/.