Spatio-Temporal Gene Discovery For Autism Spectrum Disorder

Abstract

Abstract-Large-scale whole exome sequencing studies for Autism Spectrum Disorder (ASD) could identify only around six dozen risk genes because the genetic architecture of the disorder is highly complex, with roughly a thousand genes involved. To speed the gene discovery process up, a few network-based ASD gene discovery algorithms were proposed in the literature, which assume ASD risk genes are working as a functional cluster. Even though all these methods use static gene interaction networks, the functional clustering of genes is bound to evolve during brain development and any disruptions are likely to have a cascading effect on the future associations. Thus, any approach that disregards the dynamic nature of neurodevelopment is limited in its power to detect ASD risk genes. While the assumption of a clustering of ASD genes is sensible, we hypothesize that the clustering is not necessarily static, but rather dynamic and spatiotemporal. Here, we present a spatio-temporal gene discovery algorithm for ASD, which leverages information from evolving gene coexpression networks of neurodevelopment. The algorithm (ST-Steiner) solves a prize-collecting Steiner forest-based problem on coexpression networks as a model for brain development and transfers information from precursor neurodevelopmental windows. We applied the algorithm on exome sequencing data of 3,871 samples and identified risk clusters using gene coexpression networks of early-fetal and mid-fetal periods. On an independent dataset, we demonstrate that incorporation of temporal dimension increases the prediction power and that the predicted clusters are hit (validated) more, compared to the state-of-theart methods. We also show that our predicted clusters show higher overlap with known ASD risk genes and genes associated with known ASD-related functionalities. ST-Steiner is available at http://ciceklab.cs.bilkent.edu.tr/st-steiner

Synthesizing Signaling Pathways from Temporal Phosphoproteomic Data

“…Participants in the HPN-DREAM network inference challenge ( Hill et al, 2016 ) inferred signaling networks from time series data for tens of phosphoproteins, but the top methods either did not scale to our dataset (PropheticGranger; Carlin et al, 2017 ) or did not perform well (FunChisq; Zhang and Song, 2013 ). Other algorithms that integrate temporal information with PPI networks ( Budak et al, 2015 ; Gitter and Bar-Joseph, 2013 ; Jain et al, 2016 ; Norman and Cicek, 2018 ; Patil et al, 2013 ) do not evaluate and summarize all pathway models that are supported by the network and phosphorylation timing constraints. This summarization strategy is what enables TPS to scale to solution spaces ( Figure S6 ) that are substantially larger than those typically considered by declarative computational approaches ( Chasman et al, 2014 ; Dunn et al, 2014 ; Guziolowski et al, 2013 ; Kӧksal et al, 2013 ; Moignard et al, 2015 ; Sharan and Karp, 2013 ).…”