DeepPheno: Predicting single gene loss of function phenotypes

DeepPheno is a method for predicting gene-phenotype (HPO classes) associations from gene functional annotations (GO classes) and gene expression values.

This repository contains script which were used to build and train the DeepPheno model together with the scripts for evaluating the model's performance.

Dependencies

• The code was developed and tested using python 3.7.
• To install python dependencies run: pip install -r requirements.txt

Data

• https://bio2vec.cbrc.kaust.edu.sa/data/deeppheno/ - Here you can find the data used to train and evaluate our method.
• data.tar.gz - data folder with latest dataset
• data-cafa2.tar.gz - CAFA2 challenge dataset
• predictions.txt.gz - DeepPheno predictions for human genes

Scripts

The scripts require GeneOntology and Human Phenotype Ontology in OBO Format.

• uni2pandas.py - This script is used to convert data from UniProt database format to pandas dataframe.
• data.py - This script is used to generate training and testing datasets.
• pheno.py - This script is used to train the model
• evaluate_*.py - The scripts are used to compute Fmax, Smin
• GeneDis.groovy - This script is used to compute semantic similarity between gene and disease phenotypes

Running

• Download all the files from https://bio2vec.cbrc.kaust.edu.sa/data/deeppheno/data.tar.gz and place them into data folder
• run python pheno.py to start training the model

Citation

If you use DeepPheno for your research, or incorporate our learning algorithms in your work, please cite:

Maxat Kulmanov, Robert Hoehndorf; DeepPheno: Predicting single gene knockout phenotypes. BioArxiv, https://doi.org/10.1101/839332