DeepGOPlus: Improved protein function prediction from sequence
DeepGOPlus is a novel method for predicting protein functions from protein sequences using deep neural networks combined with sequence similarity based predictions.
This repository contains script which were used to build and train the DeepGOPlus model together with the scripts for evaluating the model's performance.
The code was developed and tested using python 3.6. To install python dependencies run: pip install -r requirements.txt
- http://deepgoplus.bio2vec.net/data/ - Here you can find the data used to train and evaluate our method.
- data.tar.gz - Data required to run predict.sh script
- data-cafa.tar.gz - CAFA3 challenge dataset
- data-2016.tar.gz - Dataset which is used to compare DeepGOPlus with GOLabeler and DeepText2GO
The scripts require GeneOntology in OBO Format.
- uni2pandas.py - This script is used to convert data from UniProt database format to pandas dataframe.
- deepgoplus_data.py - This script is used to generate training and testing datasets.
- deepgoplus.py - This script is used to train the model
- evaluate_*.py - The scripts are used to compute Fmax, Smin and AUPR
- Download all the files from http://deepgoplus.bio2vec.net/data/data.tar.gz and place them into data folder
- Install diamond program on your system (diamond command should be available)
- run predict.sh <input_fasta_filename> <results_filename>
The online version of DeepGOPlus is available at http://deepgoplus.bio2vec.net/
If you use DeepGOPlus for your research, or incorporate our learning algorithms in your work, please cite: Maxat Kulmanov, Robert Hoehndorf; DeepGOPlus: Improved protein function prediction from sequence, Bioarxiv, https://www.biorxiv.org/content/10.1101/615260v1