Datasets for DyProL: Conformational Ensembles of Nucleic Acid-Binding Proteins

This directory contains the benchmark datasets and structural resources used by DyProL for protein-nucleic acid binding-site prediction. The full dataset package should be placed at the root of the DyProL repository as Datasets/ before running the code.

Overview

DyProL evaluates two benchmark settings:

DyProL/: the primary dynamic-ensemble dataset used in this work.

GraphBind/: the public benchmark dataset used for cross-dataset evaluation and comparison with prior methods.

Each benchmark is organized by ligand type:

DNA/

RNA/

The DNA-binding protein (DBP) subset of the DyProL dataset contains:

DNA-1022_Train.txt: list of 1,022 DNA-binding proteins in the training set.

DNA-256_Test.txt: list of 256 DNA-binding proteins in the test set.

PDB/: experimentally determined or reference structural inputs.

BioEmu/: conformational ensembles generated by BioEmu for dynamic modeling.

AlphaFlow/, ESMFlow/, ESMDiff/: alternative conformational ensembles used for ensemble-quality comparison.

The RNA-binding protein (RBP) subset of the DyProL dataset contains:

RNA-932_Train.txt: list of 932 RNA-binding proteins in the training set.

RNA-234_Test.txt: list of 234 RNA-binding proteins in the test set.

PDB/: experimentally determined or reference structural inputs.

BioEmu/: conformational ensembles generated by BioEmu for dynamic modeling.

The DNA-binding protein (DBP) subset of the GraphBind dataset contains:

DNA-573_Train.txt: list of 573 DNA-binding proteins in the training set.

DNA-129_Test.txt: list of 129 DNA-binding proteins in the test set.

BioEmu/: conformational ensembles generated by BioEmu for dynamic modeling.

ESM_MSA_PSSM/: sequence-derived features used in the GraphBind benchmark, including MSA, PSSM, and related embedding files.

The RNA-binding protein (RBP) subset of the GraphBind dataset contains:

RNA-495_Train.txt: list of 495 RNA-binding proteins in the training set.

RNA-117_Test.txt: list of 117 RNA-binding proteins in the test set.

BioEmu/: conformational ensembles generated by BioEmu for dynamic modeling.

ESM_MSA_PSSM/: sequence-derived features used in the GraphBind benchmark, including MSA, PSSM, and related embedding files.

Directory Structure

Datasets/
├── DyProL/
│ ├── DNA/
│ │ ├── DNA-1022_Train.txt
│ │ ├── DNA-256_Test.txt
│ │ ├── PDB/
│ │ ├── BioEmu/
│ │ ├── AlphaFlow/
│ │ ├── ESMFlow/
│ │ └── ESMDiff/
│ └── RNA/
│ ├── RNA-932_Train.txt
│ ├── RNA-234_Test.txt
│ ├── PDB/
│ └── BioEmu/
└── GraphBind/
├── DNA/
│ ├── DNA-573_Train.txt
│ ├── DNA-129_Test.txt
│ ├── BioEmu/
│ └── ESM_MSA_PSSM/
└── RNA/
├── RNA-495_Train.txt
├── RNA-117_Test.txt
├── BioEmu/
└── ESM_MSA_PSSM/

File Content

*_Train.txt and *_Test.txt list the proteins assigned to each split.

PDB/ contains structural inputs for static baselines and controlled comparisons.

BioEmu/ stores sampled equilibrium-like conformational ensembles used by DyProL.

AlphaFlow/, ESMFlow/, and ESMDiff/ provide additional sampled ensembles for ensemble-quality benchmarking.

ESM_MSA_PSSM/ stores sequence-derived features and structural auxiliary inputs required by GraphBind-style baselines, which is curated from https://zenodo.org/records/10120954.

Usage Notes

The code expects the Datasets/ directory to remain at the project root.

After downloading the Zenodo package, unzip it and place the resulting Datasets/ folder in the repository root so that all relative paths used by the code resolve correctly.