Predictive Genomics leveraging Multi-Task learning and LINA
- Python 3.6
- Tensoflow v2.6.2
- pandas_plink v2.2.9
- pandas v1.4.2
- numpy v1.19.5
- scikit-learn v1.1.0
This github contains the base code to run our experimentations. Each module represent a step of analysis.
- Data_Preproccessing
- phenotype_extraction_code.ipynb:Contain the code for the parralel phenotypes extraction. Produce 2 CSV file containing cancers and non-cancer diseases
- Models
- model_STL.py:Contain the code to train one the STL model. Require the phenotype index number as argument
- Launch_trainings_STL.sh and train_stl.sbatch: Scripts to submit 1 job per phenotype to train STL models using SLURM.
- model_pan_disease.ipynb: Contains the code to train the pan-disease model
- model_pan_cancer.ipynb :Contains the code to train the pan-cancer model
- model_pan_cancer_interpretation_decoy.ipynb: Contains the code to train the pan-cancer model with decoy SNPs and to interpret it.
- Analyses:
- important_snps_analysis.ipynb:Compute the important snps at different thresholds of FDR
- important_snps_manhattan_venn_inter_corr_union.ipynb:compute the different metrics on the important SNPs and venn diagrams
For any inquiry please contact adrien.f.badre-1@ou.edu