Machine Learning–Based Gene Prioritization Identifies Novel Candidate Risk Genes for Inflammatory Bowel Disease

Abstract

Our method successfully differentiated IBD-risk genes from non-IBD genes by using information from expression data and a multitude of gene annotations. Crucial features were defined, and we were able to detect novel candidate risk genes for IBD. These findings may help detect new IBD-risk genes and improve the understanding of IBD pathogenesis.

Reaching the End-Game for GWAS: Machine Learning Approaches for the Prioritization of Complex Disease Loci

“…In a logistic regression, it is typical to apply a regularization terme.g., L1 (the sum of the absolute value of feature weights) and L2 (the sum of squared feature weights) -that introduce some bias while reducing variance, thereby improving predictive ability (Demir-Kavuk et al, 2011). Isakov et al (2017) used elastic net logistic regression (Zou and Hastie, 2005) which combines L1 and L2 penalties to prioritize IBD genes. This method performs both variable selection (L1), and shrinks coefficient sizes to reduce variance (L2) (Ogutu et al, 2012).…”

“…Regularized logistic regression with elastic net aims to minimize the "curse of dimensionality"where data has a larger number of features than samples -which is a particular blight on GWAS. For example, Isakov et al (2017) used data consisting of 314 positive genes and 1,736 negative genes each annotated with 1,027 features. By applying logistic regression with elastic net they could then select the best data for their models (309 features selected which are predominantly from biological ontologies).…”

“…They found that logistic regression was the highest performing ML model -emphasizing that a classification problem may require simpler solutions. Besides ensemble learning and logistic regression, SVM is also consistently used within studies performing benchmark comparisons (Roshan et al, 2011;Isakov et al, 2017;Maciukiewicz et al, 2018;Vitsios and Petrovski, 2019). SVM aims to plot a decision boundary between groups by measuring hyperplanes -based on the distances between the most extreme samples of each classification group (Smola and Scholkopf, 2004; Figure 2).…”

“…Overall, there is a need for benchmarking in order to select the model best suited to the data, and for post-GWAS prioritization the optimal model currently varies across diseases without a onesize-fits-all winner. An optimal model also hinges on data size and quality for reliability and performance, with studies varying in data size and choice of features -from using hundreds of selected features (Isakov et al, 2017) to others exploring tens of thousands (Deo et al, 2014). Further in silico methods need to address these aspects of ML, the lack of functionally validated associated genes at the disposal of ML, and how features are used in order to build a model tailored to post-GWAS prioritization.…”

High circulating elafin levels are associated with Crohn’s disease-associated intestinal strictures

“…We noticed that some of the non-stricturing CD patients also had high circulating elafin levels, leading to moderate accuracy when elafin alone was used in identifying stricturing CD patients. Elafin alone is not enough to indicate intestinal strictures accurately because the complexity of many clinical characteristics of the patients has not been considered (Isakov, Dotan et al, 2017, Waljee, Lipson et al, 2017.…”