An interpretable low-complexity machine learning framework for robust exome-based in-silico diagnosis of Crohn’s disease patients

Abstract

Whole exome sequencing (WES) data are allowing researchers to pinpoint the causes of many Mendelian disorders. In time, sequencing data will be crucial to solve the genome interpretation puzzle, which aims at uncovering the genotype-to-phenotype relationship, but for the moment many conceptual and technical problems need to be addressed. In particular, very few attempts at the in-silico diagnosis of oligo-to-polygenic disorders have been made so far, due to the complexity of the challenge, the relative scarcity of the data and issues such as batch effects and data heterogeneity, which are confounder factors for machine learning (ML) methods. Here, we propose a method for the exome-based in-silico diagnosis of Crohn’s disease (CD) patients which addresses many of the current methodological issues. First, we devise a rational ML-friendly feature representation for WES data based on the gene mutational burden concept, which is suitable for small sample sizes datasets. Second, we propose a Neural Network (NN) with parameter tying and heavy regularization, in order to limit its complexity and thus the risk of over-fitting. We trained and tested our NN on 3 CD case-controls datasets, comparing the performance with the participants of previous CAGI challenges. We show that, notwithstanding the limited NN complexity, it outperforms the previous approaches. Moreover, we interpret the NN predictions by analyzing the learned patterns at the variant and gene level and investigating the decision process leading to each prediction.

Machine Learning Modeling from Omics Data as Prospective Tool for Improvement of Inflammatory Bowel Disease Diagnosis and Clinical Classifications

“…The most frequently employed ML methods included penalized regression models, random forest, support vector machines, Bayesian approach and neural networks ( Table 3 ). Even though ML models are often “black boxes”, they could be used for identifying potentially causal molecular patterns of IBD by evaluating the most significant genes/features selected during the process of model training [ 52 , 53 , 62 , 63 , 64 ].…”

“…However, as the price of the NGS and other high-throughput techniques decreases over time, more frequent application of ML using omics data in IBD risk predictions is expected. Still, beside the greater availability of the high-throughput techniques, achieving good predictive results is often limited due to widespread presence of confounding effects, relatively low prevalence of IBD and high heterogeneity of the disease phenotypes [ 63 ]. These issues often limit the analyzed sample size or make the dataset less uniform.…”

“…In a recent study by Raimondi et al, the authors designed a novel neural network approach model, CDkoma, to classify CD from healthy controls using CAGI 2, 3 and 4 editions exome data [ 63 ]. Initially, the established CD associations were used for selection of predictor genes using PhenoPedia [ 73 ].…”

“…For each gene in the set, the gene function score was computed on the basis of predicted functional effects of all its variants. This scoring system was far better compared to the one that calculated the total number of risk variants per gene [ 63 , 64 ]. Wang analyzed the performance of SVM model with leave-one-out cross-validation on CAGI4 as CD-train and CAGI3 as CD-test dataset.…”

Decoding the effects of synonymous variants

“…A modification of polygenic risk scoring methods (111,112) to only account for effect variants may represent one approach, although it would be limited by the location of most GWAS SNPs in non-coding regions. Another approach is to unite only the coding variant effects by aggregating all variants per gene to predict disease predisposition (e.g., (113,114)) synVep predictions (as well as those of other predictors) may be plugged in these pipelines to explore the contribution of sSNVs to complex diseases.…”