PeptideMTR: MLM Variant

This model is part of the PeptideMTR suite, to be released in a forthcoming paper. It is designed for multi-target biochemical property prediction.

Overview

• Architecture: BERT-based Transformer
• Pretraining: MLM
• Input: SMILES string

All models in this suite are designed for simplicity. For inference, the model accepts a SMILES string directly.

Usage

Example scripts can be found at https://github.com/AaronFeller/PeptideMTR

Dataset

The training data and pre-computed 99 RDKit descriptors are available at PeptideMTR_training_data.

Citation

Please cite the forthcoming PeptideMTR paper for methodology and benchmark results.