The PLAS-HES-5k dataset is a curated collection of 5,000 protein-ligand complexes (PLCs) designed specifically for machine learning applications in drug discovery. This comprehensive dataset includes both bound and unbound conformations, annotated with binding free energies derived from non-equilibrium molecular dynamics simulations and approximate free energy calculations. The structural and energetic diversity represented in PLAS-HES-5k makes it an ideal benchmark and training resource for predictive and generative ML/DL models aimed at improving drug-target binding predictions.

The dataset contains:

• 5,000 protein-ligand complexes in two conformational variants:

  • PLAS: Bound conformations starting from PDB database
  • HES: Synthetic unbound conformations derived from PLAS-20K, including both low and high energy states

• For each complex:

  • Atomic coordinates
  • Binding free energy data
  • Parameter files for simulation

Each protein-ligand complex in the dataset includes:

• Bound conformations from PLAS-20K
• Complete atomic coordinates in PDB format
• Parameter files (.prmtop)
• Binding affinity calculations

• Synthetic unbound conformations seeded from PLAS-20K
• Both low and high energy conformational states
• Complete atomic coordinates in PDB format
• Parameter files (.prmtop)
• Binding affinity calculations

• .tar.gz archives containing collections of PLCs
• .txt files listing which PLCs are present in each archive
• .pdb files containing structural information
• .prmtop files containing molecular dynamics parameters
• .csv files with binding affinity data

The PLAS-HES-5k dataset is designed for:

1. Training ML/DL models for predicting protein-ligand binding affinities
2. Evaluating generative models in drug design
3. Research on conformational sampling and binding dynamics
4. Benchmarking novel ML/DL architectures for drug discovery applications

1. Environment Setup

   • Environment configuration files for simulations are located in the Env folder
   • A YAML file is provided for setting up the conda environment with Plumed2 activated

2. Steered Molecular Dynamics

   • Simulation scripts are available in the Steered_Molecular_Dynamics folder
   • Submit jobs using the command:

     sh Complete_simulation_setup_hs.sh "Index_Number" "PDB_ID" "cpu-number" "partition"
     

3. Trajectory Validation

   • The Trajectory_validation folder contains scripts to validate trajectories through:
     • Sigmoidal curve fitting
     • RMSD analysis of protein and ligand
     • Center-of-mass distance separation measurements

4. Dataset Access

   • File structures are generated for each PDB ID
   • The complete dataset is publicly available on the India-Data website: https://india-data.org/dataset-details/ef3a1c5b-6ff2-49f7-ae7a-a99f69003849
   • Extract the .tar.gz archives to access individual PLC data

5. Energy Component Analysis

   • The Distribution_Of_Energy_Components folder contains scripts for:
     • Reproducing energy component distributions across all PLCs
     • Analyzing energy components for individual PLCs

6. Training Machine Learning Models

   • The dataset is suitable for various ML/DL approaches:
     • Graph Neural Networks
     • 3D Convolutional Networks
     • Equivariant Neural Networks
     • Attention-based models

7. Benchmarking

   • Use the binding affinity data to evaluate model performance
   • Compare model predictions across both PLAS (bound) and HES (unbound) conformations

If you use this dataset in your research, please cite: [Citation information to be provided by dataset creators after the publication of the dataset]

If you have any query regarding the dataset, you can reach out to Prathit Chatterjee (prathit.chatterjee@ihub-data.iiit.ac.in).

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

1.0.0