Using game theory and decision decomposition to effectively discern and characterise bi-locus diseases

Abstract

In order to gain insight into oligogenic disorders, understanding those involving bi-locus variant combinations appears to be key. In prior work, we showed that features at multiple biological scales can already be used to discriminate among two types, i.e. disorders involving true digenic and modifier combinations. The current study expands this machine learning work towards dual molecular diagnosis cases, providing a classifier able to effectively distinguish between these three types. To reach this goal and gain an in-depth understanding of the decision process, game theory and tree decomposition techniques are applied to random forest predictors to investigate the relevance of feature combinations in the prediction. A machine learning model with high discrimination capabilities was developed, effectively differentiating the three classes in a biologically meaningful manner. Combining prediction interpretation and statistical analysis, we propose a biologically meaningful characterization of each class relying on specific feature strengths. Figuring out how biological characteristics shift samples towards one of three classes provides clinically relevant insight into the underlying biological processes as well as the disease itself.

An accurate prediction model of digenic interaction for estimating pathogenic gene pairs of human diseases

“…● Discriminate dual molecular diagnoses from digenic interaction effect. Multiple (or dual) molecular diagnoses (DD) refers to the conjunction of two independent diseases caused by different mutated genes that show simultaneously on one patient [63] . DD is a type of bi-locus effect but not the digenic interaction effect, which is often confounded with digenic interaction in practice.…”

“…However, DD is a completely different concept, which means the coincidental independent segregation of two separate disease entities, and each one is caused by variants in separate linked or unlinked genes/loci [7] . Namely, DD refers to the conjunction of two independent monogenic diseases that show simultaneously in one patient [63] , indicating that two distinct genes lead to different disease phenotypes. So genetic loci involved in DD segregate independently in most instances [55] .…”

The GENDULF algorithm: mining transcriptomics to uncover modifier genes for monogenic diseases

“…The relevant literature includes instances of a GCD with a modifier, instances of true digenic inheritance where neither gene alone is sufficient to cause the full disease, and other more complicated two‐gene scenarios. When there are sufficiently many examples, these three scenarios can be formally distinguished with machine learning techniques (Gazzo et al , 2017; Versbraegen et al , 2019). We identified seven sets of diseases that have so far received considerable attention in such studies: CF, SMA, globinopathies (thalassemias, sickle‐cell anemia), deafness, long QT syndrome, ciliopathies especially Bardet–Biedl syndrome, and hypogonadotropic hypogonadism (including Kallman Syndrome) (Schäffer, 2013; Kousi & Katsanis, 2015; Gazzo et al , 2016).…”

Multidisciplinary In-Depth Investigation in a Young Athlete Suffering from Syncope Caused by Myocardial Bridge

“…The predicted genes/variant are ranked by Gene Damage Index (GDI) [40]. Additionally, a machine-learning based method was also employed to predict digenic effect [41,42] of a pathogenic digenic variant combination identified and ranked by pathogenicity scores and confidence intervals [38]. For exploration purposes, the predicted oligogenic information was represented with gene networks in ORVAL to explore potential protein-protein interactions (PPI).…”