Model- and data-driven approaches for genotype-phenotype data
Alex Gavryushkin
Model-based approaches to genotype-phenotype data
f(x1,…,xp)=β0+∑ixiβi+∑i≤jxixjβi,j
+∑i≤j≤kxixjxkβi,j,k+…
State of the art in 2018
Kieran Elmes et al. Learning epistatic gene interactions from perturbation screens. 2021
Problem #1: The hierarchy assumptions
Sarah Howles et al. Genetic variants of calcium and vitamin D metabolism in kidney stone disease. 2019
Problem #2: Scalability
Kieran Elmes et al. Learning epistatic gene interactions from perturbation screens. 2021
Kieran Elmes et al. A fast lasso-based method for inferring higher-order interactions
Problem #3 (unsolved!):
Model assumptions
f(x1,…,xp)=β0+∑ixiβi+∑i≤jxixjβi,j +∑i≤j≤kxixjxkβi,j,k+…Kieran Elmes et al. A fast lasso-based method for inferring higher-order interactions
SNVformer
Kieran Elmes et al. SNVformer: an attention-based deep neural network for GWAS data
"eQTL" simulations (work in progress)
- Multi-Layer Perceptron [ AI ]
- Logistic Regression [ :) ]
- Transformer-encoder [ AI ]
- LightGBM (Random Forests) [ AI? ]
- Differentiable Logic [ AI ]
- Support Vector Machines [ AI? ]
Why don't you just do AI?
What AI?
bioDS lab @UCNZ
Thank you!
Model- and data-driven approaches for genotype-phenotype data Alex Gavryushkin The 33 rd Annual Queenstown Molecular Biology Meeting (QMB 2023)