Introducing PLAPT

We’re excited to announce the release of PLAPT (Protein-Ligand Affinity Prediction with Transformers), our latest breakthrough in computational drug discovery. PLAPT achieves state-of-the-art performance in predicting protein-ligand binding affinities while requiring significantly less computational resources than existing approaches.

Key Innovations

PLAPT introduces several key innovations in protein-ligand binding prediction:

• Leverages pre-trained transformers (ProtBERT and ChemBERTa) for efficient sequence processing
• Novel branching neural network architecture for feature integration
• Works with simple 1D inputs (protein sequence and SMILES notation)
• Requires 5 orders of magnitude less compute for training
• Achieves comparable or better results versus existing methods

Performance Metrics

Our model demonstrates strong predictive capabilities on standard benchmarks:

Correlation between predicted and experimental binding affinities on Test2016_290 dataset

Technical Architecture

PLAPT employs a dual-branch architecture:

1. Protein Branch: Processes amino acid sequences through ProtBERT
2. Ligand Branch: Encodes SMILES strings via ChemBERTa
3. Integration: Combines features through a branching neural network
4. Prediction: Final layers output binding affinity estimates

The entire model can be trained on a single GPU in approximately 3 minutes.

Implementation

The model and code are available on our GitHub repository: github.com/Bindwell/PLAPT

Basic usage example:

from plapt import PLAPT

model = PLAPT()
affinity = model.predict(protein_sequence, smiles_string)

Impact and Future Directions

PLAPT represents a significant step toward making computational drug discovery more accessible and efficient. Its lightweight nature and strong performance make it particularly suitable for rapid iteration in drug discovery pipelines.

We welcome contributions and feedback from the research community. Visit our GitHub repository to get involved or read our accompanying paper for technical details.

Are you interested in learning more about PLAPT or exploring potential collaborations? Contact us for more information.