摘要
 

Next-generation sequencing enables efficient detection of various types of genetic variants and then assesses their association with phenotypes. Although a massive of machine learning methods have been developed, most of these algorithms just focus on high accuracy instead of interpretability. The lack of interpretability makes the machine learning methods hard to apply to medical and genetic issues. In this research, an explainable artificial intelligence (also known as interpretable machine learning) algorithm was proposed. A tree-based predictive model with the Bayesian model combination strategy was used to predict phenotype from genetic variants. The post-hoc explanative model was used to display the importance of phenotype-relevant genetic variants and their interactions. The proposed algorithm was applied to evaluate the effect of genetic variants on severe insulin resistance patients in the UK10K dataset.

---