PDNET

PDNET provides a curated dataset and computational scripts to train deep learning models that predict inter-residue distances for protein structure modeling.

Key Features:

• Dataset (DeepCov-derived): A curated dataset derived from the DeepCov dataset containing 3,456 representative protein chains for training and validating inter-residue distance prediction models.
• Data curation scripts: Scripts for dataset curation, input feature generation, and creation of distance maps are provided to reproduce the input data used for model development.
• Model training scripts: Pre-configured scripts for training, validation, and testing of deep learning models are included to support model development and evaluation.
• Predictive targets: Support for regression models that predict residue contacts, distance intervals, and real-valued inter-residue distances in angstroms (Å).

Scientific Applications:

• Method development: Development and benchmarking of deep learning methods for predicting inter-residue distances.
• Protein structure modeling: Use of predicted inter-residue distances to assist construction of three-dimensional protein models relevant to biological function and therapeutic design.

Methodology:

Methods explicitly include dataset derivation from DeepCov, dataset curation, input feature generation, creation of distance maps, and training, validation, and testing of deep learning regression models for contacts, distance intervals, and real-valued inter-residue distances (Å).