Identifying potential therapeutics for neurodevelopmental disorders using artificial intelligence and transcriptional data - Project Summary/Abstract Many debilitating neurodevelopmental disorders (NDDs) are caused by a deficiency or excess of a specific gene product. Most of these disorders are treated at the symptomatic level, without targeting the underlying cause. Many efforts have been made to systematically find drugs that can modulate the responsible gene product and correct the imbalance. Artificial intelligence (AI) techniques, fueled by the vast amount of publicly available data, can prioritize the most promising treatments to test. There are tens of thousands of RNA-sequencing (RNA-seq) studies from the Gene Expression Omnibus (GEO) that give a quantitative readout of many gene and drug perturbations. If a chemical changes a cell’s transcriptome similarly or oppositely to the disruption of a specific gene, the drug might support or counteract the gene’s function. Large-scale analyses of discrete pathway data and transcriptomic studies have been used to discover novel protein regulators, and even successful NDD therapeutics. However, their ability to expedite screens for NDD therapies has not been quantitatively tested. This project will use AI to curate and analyze the literature and RNA-seq data, and predict the most likely genetic and pharmacologic regulators of single genes. Aim 1 will assess whether a fully automated prediction pipeline effectively prioritizes regulators of methyl-CpG- binding protein 2 (MeCP2). MeCP2 is an ideal target upon which to systematically validate predictions. It is a dosage-sensitive protein that is widely studied and perturbed. Patients with MeCP2 deficiency (Rett syndrome) and excess (MECP2 duplication syndrome) are given symptomatic therapies, and treatments for its imbalance are being developed, but there is no definitive remedy, and affected individuals continue to die in childhood or live with cognitive and motor defects. Drugs that correct the imbalance could help affected individuals, especially if given in their infancy, before all symptoms emerge and set in. This aim will involve using neuronal lines from an individual with Rett syndrome to test the top 20 predictions of MeCP2-regulating genes and drugs. Testing the top treatments without any manual filtering will determine whether this predictor effectively prioritizes successful MeCP2 regulators. Additionally, parts of a disease signature might be pathogenic, and others protective. There thus remains a need to find druggable subnetworks within a gene disruption signature. Aim 2 will establish a network analysis that can find separate network modules surrounding a given gene, and drugs that specifically target those modules. This study will identify promising treatments that can be repurposed to rescue developmental defects in individuals with Rett syndrome, as well as genes that serve as upstream regulators of MeCP2. On a larger scale, it will quantitatively assess the ability of AI-guided predictions to expedite the search for undiscovered drug-gene relationships. This will save a tremendous amount of time and resources in the search for therapeutics for monogenic NDDs, and pave the way to making more of these disorders treatable.
