Identifying SNP associations and predicting disease risk from Genome-wide association studies using LassoNet

In this paper, we show that under certain conditions LassoNet [1] should outperform threshold of significance (P-value) methods for identifying multi-SNP disease associations and predicting disease risk using data from Genome Wide Association Studies. To demonstrate this, we built a genotype-phenotype simulation to comprehensively benchmark each methods performance in variant selection and in predicting disease risk. Our results suggest that LassoNet, with its ease of implementation, should be added to the biomedical informaticians toolkit. We release code to replicate our results.