Skip to content

Latest commit

 

History

History
78 lines (59 loc) · 2.4 KB

README.md

File metadata and controls

78 lines (59 loc) · 2.4 KB

PathoGN : Pathogenicity prediction model with a graph neural network

Example Usage

Run the demo with Varibench datasets

Varibench is a benchmark datasets for prediction of genomic variant effects.

Nair PS, Vihinen M. VariBench: A Benchmark Database for Variations. Hum Mutat. 2013, 34(1):42-9.

http://structure.bmc.lu.se/VariBench/

0. Install kGCN (https://github.com/clinfo/kGCN) with conda

$ conda create -n kgcn python=3.7 conda=4.9.2
$ conda activate kgcn
$ conda install tensorflow=1.15 joblib numpy scipy scikit-learn matplotlib pandas
$ pip install --upgrade git+https://github.com/clinfo/kGCN.git
  • currently kGCN does NOT support TensorFlow 2

1. Get Valibench datasets

sh get_dataset.sh

2. Preprocessing data

sh build_dataset.sh

3. Get and preprocess Reactome data

sh make_reactome_data.sh
python script/preprocess_reactome.py

4. make input data for GCN

sh make_data.sh

5. Run GCN and evaluate with cross validation

sh run_gcn.sh

Prediction result

score1-5.csv will be created in result, and GCN-Score will be calculated for each of them.

The correspondence between numbers and data sets is as follows：

  1. exovar_filtered_tool_scores
  2. humvar_filtered_tool_scores
  3. predictSNP_selected_tool_scores
  4. swissvar_selected_tool_scores
  5. varibench_selected_tool_scores

The prediction scores for ClinVar 20200210 datasets

PathoGN predicted the pathogenicity for all variants that were not annotated as either pathogenic or benign in the 2020 ClinVar dataset. The model was trained using the labeled data (pathogenic=10,877, benign=7504) and then used to make predictions for the 12,520 unlabeled variants.

The prediction result: PredictionResults_ConflictVariant_ClinVar20200210.tsv

These scores are also available on MGeND (https://mgend.med.kyoto-u.ac.jp/).

Reference

@article {PathoGN,
        author = {Kamada, Mayumi and Takagi, Atsuko and Kojima, Ryosuke and Tanaka, Yoshihisa and Nakatsui, Masahiko and Tanabe, Noriko and Hirata, Makoto and Yoshida, Teruhiko and Okuno, Yasushi},
        title = {Network-based pathogenicity prediction for variants of uncertain significance},
        year = {2021},
        doi = {10.1101/2021.07.15.452566},
        eprint = {https://www.biorxiv.org/content/early/2021/07/16/2021.07.15.452566.full.pdf},
        journal = {bioRxiv}
}