DeepECA

DeepECA predicts protein residue-residue contacts and related structural properties from amino acid sequences using deep neural networks (DNNs) and multiple sequence alignments (MSAs) to improve tertiary structure modeling.

Key Features:

• End-to-End DNN Prediction: Uses an end-to-end deep neural network framework to predict contact residues and secondary structures directly from amino acid sequences.
• MSA Handling: Manages both deep and shallow MSAs, assigning weights to sequences in deep MSAs to enhance precision and incorporating additional sequential features in shallow MSAs to improve long-range contact prediction.
• Multi-Task Model Integration: Integrates secondary structure prediction and solvent-accessible surface area estimation into a multi-task model.
• Ensemble Averaging: Applies ensemble averaging to refine predictions and demonstrated superior or equivalent results compared to existing meta-predictors on past CASP target protein domains.
• Noise Reduction from Sequence Data: Reduces noise in abundant sequence data, including metagenomic sequences, to prevent degradation of prediction outcomes.
• Improved Tertiary Structure Prediction: Uses predicted contacts and secondary structures to generate more accurate three-dimensional models compared to evolutionary coupling analysis (ECA) methods.

Scientific Applications:

• Protein Tertiary Structure Modeling: Provides contact and secondary structure inputs to produce more accurate tertiary structural models than ECA-based approaches.
• Functional Inference and Interaction Mapping: Enables inference of protein function and interaction interfaces through high-precision contact predictions.
• Protein Design and Engineering: Supports the design of novel proteins by supplying structural constraints from predicted contacts, secondary structures, and solvent-accessible surface areas.

Methodology:

DeepECA trains deep neural network models on amino acid sequences and MSAs, assigns sequence weights for deep MSAs, incorporates additional sequential features for shallow MSAs, integrates secondary structure and solvent-accessible surface area prediction in a multi-task model, and applies ensemble averaging.