DEEPCON

DEEPCON predicts inter-residue contacts from protein sequences using dilated convolutional neural network architectures to support protein structure prediction.

Key Features:

• Convolutional Neural Network architectures: Implements advanced ConvNet architectures trained for contact prediction from protein sequence data.
• Dilated Convolutions: Expands the receptive field of the network to capture long-range dependencies in protein sequences.
• High Precision Predictions: Reports a 15% increase in precision for L/2 long-range contacts on the DeepCov dataset and outperforms ensembled DNCON2 by 4.8% for top L long-range contacts on the CASP12 dataset using a single network.

Scientific Applications:

• Structural Bioinformatics: Enables protein structure prediction and modeling by inferring three-dimensional contacts from amino acid sequences to support studies of protein function, interactions, and drug discovery.

Methodology:

Trains convolutional neural network architectures with dilated convolutions on large homologous protein sequence datasets and optimizes precision for medium-range and long-range contacts through testing and refinement of the architecture.