TransFew

Improving protein function prediction by learning and integrating representations of protein sequences and function labels

TransFew leaverages representations of both protein sequences and function labels (Gene Ontology (GO) terms) to predict the function of proteins. It improves the accuracy of predicting both common and rare function terms (GO terms).

Installation

# clone project
git clone https://github.com/BioinfoMachineLearning/TransFew.git
cd TransFew/
# download trained models and test sample
https://calla.rnet.missouri.edu/rnaminer/tfew/TFewDataset
# Unzip Dataset
unzip TFewDataset
# create conda environment
conda env create -f transfew.yaml
conda activate transfew

Prediction

Predict protein functions with TransFew
options:
-h, --help show this help message and exit
--data-path DATA_PATH Path to data files (models)
--working-dir WORKING_DIR Path to generate temporary files
--ontology ONTOLOGY Path to data files
--no-cuda NO_CUDA Disables CUDA training.
--batch-size BATCH_SIZE Batch size.
--fasta-path FASTA_PATH Path to Fasta
--output OUTPUT File to save output

4. An example of predicting cellular component of some proteins:

1. Change ROOT_DIR in CONSTANTS.py to path of data directory
2. python predict.py --data-path /TFewData/ --fasta-path output_dir/test_fasta.fasta --ontology cc --working-dir output_dir --output result.tsv

Output format

 protein GO term score
A0A7I2V2M2	GO:0043227	0.996
A0A7I2V2M2	GO:0043226	0.996
A0A7I2V2M2	GO:0005737	0.926
A0A7I2V2M2	GO:0043233	0.924
A0A7I2V2M2	GO:0031974	0.913
A0A7I2V2M2	GO:0070013	0.912
A0A7I2V2M2	GO:0031981	0.831
A0A7I2V2M2	GO:0005654	0.767

Dataset

See DATASET.md (https://github.com/BioinfoMachineLearning/TransFew/blob/main/DATASET.md) for description of data

Training

The training program is available in training.py, to train the model:

 1. Change ROOT_DIR in CONSTANTS.py to path of data directory
2. Run: python training.py

Reference

Boadu, F., & Cheng, J. (2024). Improving protein function prediction by learning and integrating representations of protein sequences and function labels. Bioinformatics Advances. Volume 4, Issue 1, vbae120.