Model description

MHC-I-EpiPred (MHC-I-EpiPred, MHC I molecular epitope prediction) is a protein language model fine-tuned from ESM2 pretrained model (facebook/esm2_t33_650M_UR50D) on a T cell MHC I epitope dataset.

MHC-I-EpiPred is is a classification model for predicting the class of MHC I epitope.

Dataset

The original data was downloaded from IEDB data base at https://www.iedb.org/home_v3.php. The full data can be downloaded at https://www.iedb.org/downloader.php?file_name=doc/tcell_full_v3.zip
This dataset comprises 543,717 T-cell epitope entries, spanning a variety of species and infections caused by diverse viruses. The epitope information included encompasses a broad range of potential sources, including data relevant to disease immunotherapy.

Finally, the dataset we used to train the model contains 41,060 positive and negative samples, which is stored in https://github.com/pengsihua2023/MHC-I-EpiPred/tree/main/data.

Results

MHC-I-EpiPred achieved the following results: Training Loss (cross-entropy loss, CEL): 0.1044
Training Accuracy: 98.99%
Evaluation Loss (cross-entropy loss, CEL): 0.1576
Evaluation Accuracy: 97.04%
Epochs: 492

Model training code at GitHub

https://github.com/pengsihua2023/MHC-I-EpiPred-ESM2

How to use MHC-I-EpiPred

An example

Pytorch and transformers libraries should be installed in your system.

Install pytorch

pip install torch torchvision torchaudio

Install transformers

pip install transformers

Run the following code

Coming soon!

Funding

This project was funded by the CDC to Justin Bahl (BAA 75D301-21-R-71738).

Model architecture, coding and implementation

Sihua Peng

sihuapeng
/

MHC-I-TCEpiPred