• Applications of sequential learning and recommender systems to pharmaceutics. • Review of Machine Learning methods in drug discovery, testing and repurposing. • Survey of available genomic data and feature selection methods for drug development.

Impairment of cognitive function is a common feature of many neurodevelopmental disorders. Systems genetics analysis in the brain uncovered a convergent gene network for both cognition and neurodevelopmental disorders. As the network does not recapitulate known pathways, this finding represents a new basis for understanding factors influencing normal and disordered cognition. Genetic determinants of cognition are poorly characterized, and their relationship to genes that confer risk for neurodevelopmental disease is unclear. Here we performed a systems-level analysis of genome-wide gene expression data to infer gene-regulatory networks conserved across species and brain regions. Two of these networks, M1 and M3, showed replicable enrichment for common genetic variants underlying healthy human cognitive abilities, including memory. Using exome sequence data from 6,871 trios, we found that M3 genes were also enriched for mutations ascertained from patients with neurodevelopmental disease generally, and intellectual disability and epileptic encephalopathy in particular. M3 consists of 150 genes whose expression is tightly developmentally regulated, but which are collectively poorly annotated for known functional pathways. These results illustrate how systems-level analyses can reveal previously unappreciated relationships between neurodevelopmental disease–associated genes in the developed human brain, and provide empirical support for a convergent gene-regulatory network influencing cognition and neurodevelopmental disease.

SOX5 encodes a transcription factor involved in the regulation of chondrogenesis and the development of the nervous system. Despite its important developmental roles, SOX5 disruption has yet to be associated with human disease. We report one individual with a reciprocal translocation breakpoint within SOX5, eight individuals with intragenic SOX5 deletions (four are apparently de novo and one inherited from an affected parent), and seven individuals with larger 12p12 deletions encompassing SOX5. Common features in these subjects include prominent speech delay, intellectual disability, behavior abnormalities, and dysmorphic features. The phenotypic impact of the deletions may depend on the location of the deletion and, consequently, which of the three major SOX5 protein isoforms are affected. One intragenic deletion, involving only untranslated exons, was present in a more mildly affected subject, was inherited from a healthy parent and grandparent, and is similar to a deletion found in a control cohort. Therefore, some intragenic SOX5 deletions may have minimal phenotypic effect. Based on the location of the deletions in the subjects compared to the controls, the de novo nature of most of these deletions, and the phenotypic similarities among cases, SOX5 appears to be a dosage‐sensitive, developmentally important gene. Hum Mutat 33:728–740, 2012.

The identification of drug targets is highly challenging, particularly for diseases of the brain. To address this problem, we developed and experimentally validated a general computational framework for drug target discovery that combines gene regulatory information with causal reasoning (“Causal Reasoning Analytical Framework for Target discovery”—CRAFT). Using a systems genetics approach and starting from gene expression data from the target tissue, CRAFT provides a predictive framework for identifying cell membrane receptors with a direction-specified influence over disease-related gene expression profiles. As proof of concept, we applied CRAFT to epilepsy and predicted the tyrosine kinase receptor Csf1R as a potential therapeutic target. The predicted effect of Csf1R blockade in attenuating epilepsy seizures was validated in three pre-clinical models of epilepsy. These results highlight CRAFT as a systems-level framework for target discovery and suggest Csf1R blockade as a novel therapeutic strategy in epilepsy. CRAFT is applicable to disease settings other than epilepsy.

The relationship between monogenic and polygenic forms of epilepsy is poorly understood and the extent to which the genetic and acquired epilepsies share common pathways is unclear. Here, we use an integrated systems-level analysis of brain gene expression data to identify molecular networks disrupted in epilepsy. We identified a co-expression network of 320 genes (M30), which is significantly enriched for non-synonymous de novo mutations ascertained from patients with monogenic epilepsy and for common variants associated with polygenic epilepsy. The genes in the M30 network are expressed widely in the human brain under tight developmental control and encode physically interacting proteins involved in synaptic processes. The most highly connected proteins within the M30 network were preferentially disrupted by deleterious de novo mutations for monogenic epilepsy, in line with the centrality-lethality hypothesis. Analysis of M30 expression revealed consistent downregulation in the epileptic brain in heterogeneous forms of epilepsy including human temporal lobe epilepsy, a mouse model of acquired temporal lobe epilepsy, and a mouse model of monogenic Dravet (SCN1A) disease. These results suggest functional disruption of M30 via gene mutation or altered expression as a convergent mechanism regulating susceptibility to epilepsy broadly. Using the large collection of drug-induced gene expression data from Connectivity Map, several drugs were predicted to preferentially restore the downregulation of M30 in epilepsy toward health, most notably valproic acid, whose effect on M30 expression was replicated in neurons. Taken together, our results suggest targeting the expression of M30 as a potential new therapeutic strategy in epilepsy.

Background: Homozygous mutations in WWOX were reported in eight individuals of two families with autosomal recessive spinocerebellar ataxia type 12 and in two siblings with infantile epileptic encephalopathy (IEE), including one who deceased prior to DNA sampling. Methods: By combining array comparative genomic hybridisation, targeted Sanger sequencing and next generation sequencing, we identified five further patients from four families with IEE due to biallelic alterations of WWOX. Results: We identified eight deleterious WWOX alleles consisting in four deletions, a four base-pair frameshifting deletion, one missense and two nonsense mutations. Genotype-phenotype correlation emerges from the seven reported families. The phenotype in four patients carrying two predicted null alleles was characterised by (1) little if any psychomotor acquisitions, poor spontaneous motility and absent eye contact from birth, (2) pharmacoresistant epilepsy starting in the 1st weeks of life, (3) possible retinal degeneration, acquired microcephaly and premature death. This contrasted with the less severe autosomal recessive spinocerebellar ataxia type 12 phenotype due to hypomorphic alleles. In line with this correlation, the phenotype in two siblings carrying a null allele and a missense mutation was intermediate. Conclusions: Our results obtained by a combination of different molecular techniques undoubtedly incriminate WWOX as a gene for recessive IEE and illustrate the usefulness of high throughput data mining for the identification of genes for rare autosomal recessive disorders. The structure of the WWOX locus encompassing the FRA16D fragile site might explain why constitutive deletions are recurrently reported in genetic databases, suggesting that WWOX-related encephalopathies, although likely rare, may not be exceptional.