Bayesian tensor factorization for predicting clinical outcomes using integrated human genetics evidence

This repo contains detailed information on data and methodology used in our work presented at ICML 2022 Computational Biology workshop.

Workshop website | Extended abstract | Poster

Schematic representation of 3-rank tensor with sparse data available on genetics evidence and clinical outcome.

Data and Methods

We integrate three lines of human genetics evidence across rare disease, gene burden and common disease. We also use NLP-based classification of clinical outcomes to label clinical "failures". All the data used in this analysis are retrieved from the latest release of Open Targets platform v22.06, Data downloads

Rare disease (Mendelian phenotypes)

Data source	Positive label	Negative label
ClinGen	Definitive or Strong	Other classification
Genomics England PanelApp	Green	Amber

Gene burden (Rare variant association)

Data source	Positive label	Negative label
REGENERON	$P\le 2.18 \times {10^{−11}}$	Else
AstraZeneca PheWAS Portal	$P\le 2 \times {10^{−9}}$	Else
Genebass	$P\le 6.7 \times {10^{−7}}$	Else

Common disease (GWAS)

Data source	Positive label	Negative label
Locus-to-gene "L2G" score	Score $\ge 0.5$	Score < $0.5$

Citation

If you find this work useful, please cite it as follows:

@misc{https://doi.org/10.48550/arxiv.2207.12538,
  doi = {10.48550/ARXIV.2207.12538},
  url = {https://arxiv.org/abs/2207.12538},
  author = {Soylemez, Onuralp},
  keywords = {Machine Learning (cs.LG), Genomics (q-bio.GN), Applications (stat.AP), FOS: Computer and information sciences, FOS: Computer and information sciences, FOS: Biological sciences, FOS: Biological sciences},
  title = {Bayesian tensor factorization for predicting clinical outcomes using integrated human genetics evidence},
  publisher = {arXiv},
  year = {2022},
  copyright = {Creative Commons Attribution 4.0 International}
}