HUME: large-scale detection of causal genetic factors of adverse drug reactions

Abstract

Supplementary data are available at Bioinformatics online.

Aristotle: stratified causal discovery for omics data

“…This indicates that the number of the features passing the filtering of Aristotle is a reasonable estimate of the number of true null hypothesis. The order of magnitude of Aristotle’s p -values is similar to those reported in the guideline [ 53 ] and HUME [ 20 ] and the distribution of the p -values significantly deviates away from the straight line, i.e. null distribution).…”

“…However, this does not mean that the only possibly true relations are guideline relations. We investigate the overlap of the causal SNPs detected by Aristotle with the SNPs discussed in the guideline and HUME [ 20 ].…”

“…In each pair, one sample has received the treatment feature and the other sample has not, but they are as similar as possible with respect to the confounders. Confounders must include both the natural confounders given in input, as well as the other candidates which were statistically significant in the initial Feature Filtering step [20,32]. In our pairing approach, we assume two types of constraints.…”

“…In [ 19 ], candidate features are selected through an association rule mining algorithm and tested in a series of retrospective cohort studies to evaluate the causality of the candidates. In HUME [ 20 ], a network of co-occurrence between features and outcomes was used to determine the most likely candidate feature-outcome combinations and their potential confounders, and the causality of the candidates was evaluated by a matched pairs QED.…”

Challenges and Opportunities in Implementing Pharmacogenetic Testing in Clinical Settings

“…Machine learning, a novel application of artificial intelligence, may be an efficient way to solve complex problems with large, diverse data sources (106)(107)(108). Several machine learning models have been developed to predict therapeutic outcomes, adverse events, or both by integrating existing large-scale pharmacogenomic datasets, such as a naive Bayesian model (a classification algorithm to rank and predict gene-drug adverse reactions) (109), HUME (a multiphase algorithm to identify causal pharmacogenomic relationships in gene and drug pathways) (110), and multi-omics late integration (MOLI) (a method to improve the accuracy of drug response prediction) (111). Although the use of machine learning is still in the early stages of development for pharmacogenetic analyses, these computational and statistical techniques are likely to provide new tools for the discovery of genetic variants through improved data mining techniques.…”