A primer on the use of machine learning to distil knowledge from data in biological psychiatry.

Applications of machine learning in the biomedical sciences are growing rapidly. This growth has been spurred by diverse cross-institutional and interdisciplinary collaborations, public availability of large datasets, an increase in the accessibility of analytic routines, and the availability of powerful computing resources. With this increased access and exposure to machine learning comes a responsibility for education and a deeper understanding of its bases and bounds, borne equally by data scientists seeking to ply their analytic wares in medical research and by biomedical scientists seeking to harness such methods to glean knowledge from data. This article provides an accessible and critical review of machine learning for a biomedically informed audience, as well as its applications in psychiatry. The review covers definitions and expositions of commonly used machine learning methods, and historical trends of their use in psychiatry. We also provide a set of standards, namely Guidelines for REporting Machine Learning Investigations in Neuropsychiatry (GREMLIN), for designing and reporting studies that use machine learning as a primary data-analysis approach. Lastly, we propose the establishment of the Machine Learning in Psychiatry (MLPsych) Consortium, enumerate its objectives, and identify areas of opportunity for future applications of machine learning in biological psychiatry. This review serves as a cautiously optimistic primer on machine learning for those on the precipice as they prepare to dive into the field, either as methodological practitioners or well-informed consumers.

A polygenic resilience score moderates the genetic risk for schizophrenia: Replication in 18,090 cases and 28,114 controls from the Psychiatric Genomics Consortium.

Identifying heritable factors that moderate the genetic risk for schizophrenia (SCZ) could help clarify why some individuals remain unaffected despite having relatively high genetic liability. Previously, we developed a framework to mine genome-wide association (GWAS) data for common genetic variants that protect high-risk unaffected individuals from SCZ, leading to derivation of the first-ever "polygenic resilience score" for SCZ (resilient controls n = 3786; polygenic risk score-matched SCZ cases n = 18,619). Here, we performed a replication study to verify the moderating effect of our polygenic resilience score on SCZ risk (OR = 1.09, p = 4.03 × 10-5 ) using newly released GWAS data from 23 independent case-control studies collated by the Psychiatric Genomics Consortium (PGC) (resilient controls n = 2821; polygenic risk score-matched SCZ cases n = 5150). Additionally, we sought to optimize our polygenic resilience-scoring formula to improve subsequent modeling of resilience to SCZ and other complex disorders. We found significant replication of the polygenic resilience score, and found that strict pruning of SNPs based on linkage disequilibrium to known risk SNPs and their linked loci optimizes the performance of the polygenic resilience score.

A twin analysis to estimate genetic and environmental factors contributing to variation in weighted gene co-expression network module eigengenes.

Multivariate network-based analytic methods such as weighted gene co-expression network analysis are frequently applied to human and animal gene-expression data to estimate the first principal component of a module, or module eigengene (ME). MEs are interpreted as multivariate summaries of correlated gene-expression patterns and network connectivity across genes within a module. As such, they have the potential to elucidate the mechanisms by which molecular genomic variation contributes to individual differences in complex traits. Although increasingly used to test for associations between modules and complex traits, the genetic and environmental etiology of MEs has not been empirically established. It is unclear if, and to what degree, individual differences in blood-derived MEs reflect random variation versus familial aggregation arising from heritable or shared environmental influences. We used biometrical genetic analyses to estimate the contribution of genetic and environmental influences on MEs derived from blood lymphocytes collected on a sample of N = 661 older male twins from the Vietnam Era Twin Study of Aging (VETSA) whose mean age at assessment was 67.7 years (SD = 2.6 years, range = 62-74 years). Of the 26 detected MEs, 14 (56%) had statistically significant additive genetic variation with an average heritability of 44% (SD = 0.08, range = 35%-64%). Despite the relatively small sample size, this demonstration of significant family aggregation including estimates of heritability in 14 of the 26 MEs suggests that blood-based MEs are reliable and merit further exploration in terms of their associations with complex traits and diseases.

The genetic and environmental etiology of novel frequency-driven regional parcellations of abnormal white matter.

The prevalence of white matter disease increases with age and is associated with cerebrovascular disease, cognitive decline, and risk for dementia. MRI measures of abnormal signal in the white matter (AWM) provide estimates of damage, however, regional patterns of AWM may be differentially influenced by genetic or environmental factors. With our data-driven regional parcellation approach, we created a probability distribution atlas using Vietnam Era Twin Study of Aging (VETSA) data (n = 475, mean age 67.6 years) and applied a watershed algorithm to define separate regional parcellations. We report biometrical twin modeling for five anatomically distinct regions: (1) Posterior, (2) Superior frontal and parietal, (3) Anterior and inferior frontal with deep areas, (4) Occipital, and (5) Anterior periventricular. We tested competing multivariate hypotheses to identify unique influences and to explain sources of covariance among the parcellations. Family aggregation could be entirely explained by additive genetic influences, with additive genetic variance (heritability) ranging from 0.69 to 0.79. Most genetic correlations between parcellations ranged from moderate to high (rg = 0.57-0.85), although two were small (rg = 0.35-0.39), consistent with varying degrees of unique genetic influences. This proof-of-principle investigation demonstrated the value of our novel, data-driven parcellations, with identifiable genetic and environmental differences, for future exploration.

Novel genome-wide associations for effort valuation and psychopathology in children and adults.

The Research Domain Criteria (RDoC) initiative was established by the US National Institute of Mental Health as a multilevel, disorder-agnostic framework for analysis of human psychopathology through designated domains and constructs, including the "Positive Valence Systems" domain focused on reward-related behavior. This study investigates the reward valuation subconstruct of "effort" and its association with genetic markers, functional neurobiological pathways, and polygenic risk scores for psychopathology in 1215 children aged 6-12 and their parents (n = 1044). All participants completed the effort expenditure for rewards task (EEfRT), which assesses "effort" according to two quantitative measures: hard-task choice and reward sensitivity. Genetic association analyses were undertaken in MAGMA, utilizing EEfRT outcome variables as genome-wide association studies phenotypes to compute SNP and gene-level associations. Genome-wide association analyses found two distinct genetic loci that were significantly associated with measures of reward sensitivity and a separate genetic locus associated with hard task choice. Gene-set enrichment analysis yielded significant associations between "effort" and multiple gene sets involved in reward processing-related pathways, including dopamine receptor signaling, limbic system and forebrain development, and biological response to cocaine. These results serve to establish "effort" as a relevant construct for understanding reward-related behavior at the genetic level and support the RDoC framework for assessing disorder-agnostic psychopathology.

Autophagy, apoptosis, and neurodevelopmental genes might underlie selective brain region vulnerability in attention-deficit/hyperactivity disorder.

Large-scale brain imaging studies by the ENIGMA Consortium identified structural changes associated with attention-deficit/hyperactivity disorder (ADHD). It is not clear why some brain regions are impaired and others spared by the etiological risks for ADHD. We hypothesized that spatial variation in brain cell organization and/or pathway expression levels contribute to selective brain region vulnerability (SBRV) in ADHD. In this study, we used the largest available collection of magnetic resonance imaging (MRI) results from the ADHD ENIGMA Consortium (subcortical MRI n = 3242; cortical MRI n = 4180) along with high-resolution postmortem brain microarray data from Allen Brain Atlas (donors n = 6) from 22 brain regions to investigate our SBRV hypothesis. We performed deconvolution of the bulk transcriptomic data to determine abundances of neuronal and nonneuronal cells in the brain. We assessed the relationships between gene-set expression levels, cell abundance, and standardized effect sizes representing regional changes in brain sizes in cases of ADHD. Our analysis yielded significant correlations between apoptosis, autophagy, and neurodevelopment genes with smaller brain sizes in ADHD, along with associations to regional abundances of astrocytes and oligodendrocytes. The lack of enrichment of common genetic risk variants for ADHD within implicated gene sets suggests an environmental etiology to these differences. This work provides novel mechanistic clues about SBRV in ADHD.

A polygenic resilience score moderates the genetic risk for schizophrenia.

Based on the discovery by the Resilience Project (Chen R. et al. Nat Biotechnol 34:531-538, 2016) of rare variants that confer resistance to Mendelian disease, and protective alleles for some complex diseases, we posited the existence of genetic variants that promote resilience to highly heritable polygenic disorders1,0 such as schizophrenia. Resilience has been traditionally viewed as a psychological construct, although our use of the term resilience refers to a different construct that directly relates to the Resilience Project, namely: heritable variation that promotes resistance to disease by reducing the penetrance of risk loci, wherein resilience and risk loci operate orthogonal to one another. In this study, we established a procedure to identify unaffected individuals with relatively high polygenic risk for schizophrenia, and contrasted them with risk-matched schizophrenia cases to generate the first known "polygenic resilience score" that represents the additive contributions to SZ resistance by variants that are distinct from risk loci. The resilience score was derived from data compiled by the Psychiatric Genomics Consortium, and replicated in three independent samples. This work establishes a generalizable framework for finding resilience variants for any complex, heritable disorder.

Brief Report: Expectancies for alcohol analgesia are associated with greater alcohol use among moderate-to-heavy drinkers without chronic pain.

BACKGROUND AND OBJECTIVES: Expectancies for alcohol analgesia (i.e., expectations that drinking alcohol will reduce pain) have been associated with greater alcohol consumption among individuals with chronic pain, and there is reason to believe that such expectancies may also contribute to drinking behavior among alcohol users without a current chronic pain condition. Therefore, the objective of these analyses was to test associations between a measure of expectancies for alcohol analgesia (EAA) and alcohol use among drinkers without current pain. METHOD: These are secondary analyses of baseline data collected from 200 moderate-to-heavy adult drinkers (39% women). RESULTS: EAA scores were positively associated with quantity/frequency of drinking, urge to drink, and other alcohol outcome expectancies (ps < .01). DISCUSSION AND CONCLUSIONS: Expectancies that alcohol will reduce pain are associated with heavier drinking among drinkers without pain. Over time, such expectancies may contribute to the development of alcohol use disorder and chronically painful conditions. SCIENTIFIC SIGNIFICANCE: This study provides the first evidence that even moderate-to-heavy drinkers without chronic pain may still hold expectancies for alcohol analgesia, and that this may be related to greater quantity/frequency of drinking.

Bipolar Disorder, Religion, and Spirituality: A Scoping Review.

Mental health clinicians frequently study the religion and spirituality (R/S) of their patients. There is, however, a paucity of empirical research concerning R/S of patients with bipolar disorder. This lack is exacerbated by the absence of an evaluation of how these studies relate to each other. Reviews to date concern almost exclusively quantitative studies; a review that synthesizes quantitative and qualitative research is needed. The aim of this paper is to provide a synthesis of empirical studies that is useful in clinical practice. Systematic searches for relevant journal articles in SCOPUS, PubMed, and PsycInfo found 14 quantitative and four qualitative studies. The research reveals that intrinsic religiosity and positive religious coping are the dimensions of R/S that have the most positive correlations with improvement of bipolar disorder symptoms as revealed by measures of clinical outcomes. Patients struggle with their religious experiences, and they wish that R/S would be taken into account by mental health professionals. The quantitative studies are not in conflict with the patient/person-centered focus of qualitative studies. This integration of quantitative data with a patient/person-centered focus shows how belief and illness affect each other. The tensions inherent in such an integration provide new insights for research and treatment. Unfortunately, the qualitative literature has not caught up with quantitative approaches in terms of diagnostic rigor.

Psychiatric genetics and the structure of psychopathology.

For over a century, psychiatric disorders have been defined by expert opinion and clinical observation. The modern DSM has relied on a consensus of experts to define categorical syndromes based on clusters of symptoms and signs, and, to some extent, external validators, such as longitudinal course and response to treatment. In the absence of an established etiology, psychiatry has struggled to validate these descriptive syndromes, and to define the boundaries between disorders and between normal and pathologic variation. Recent advances in genomic research, coupled with large-scale collaborative efforts like the Psychiatric Genomics Consortium, have identified hundreds of common and rare genetic variations that contribute to a range of neuropsychiatric disorders. At the same time, they have begun to address deeper questions about the structure and classification of mental disorders: To what extent do genetic findings support or challenge our clinical nosology? Are there genetic boundaries between psychiatric and neurologic illness? Do the data support a boundary between disorder and normal variation? Is it possible to envision a nosology based on genetically informed disease mechanisms? This review provides an overview of conceptual issues and genetic findings that bear on the relationships among and boundaries between psychiatric disorders and other conditions. We highlight implications for the evolving classification of psychopathology and the challenges for clinical translation.

An integrated analysis of genes and functional pathways for aggression in human and rodent models.

Human genome-wide association studies (GWAS), transcriptome analyses of animal models, and candidate gene studies have advanced our understanding of the genetic architecture of aggressive behaviors. However, each of these methods presents unique limitations. To generate a more confident and comprehensive view of the complex genetics underlying aggression, we undertook an integrated, cross-species approach. We focused on human and rodent models to derive eight gene lists from three main categories of genetic evidence: two sets of genes identified in GWAS studies, four sets implicated by transcriptome-wide studies of rodent models, and two sets of genes with causal evidence from online Mendelian inheritance in man (OMIM) and knockout (KO) mice reports. These gene sets were evaluated for overlap and pathway enrichment to extract their similarities and differences. We identified enriched common pathways such as the G-protein coupled receptor (GPCR) signaling pathway, axon guidance, reelin signaling in neurons, and ERK/MAPK signaling. Also, individual genes were ranked based on their cumulative weights to quantify their importance as risk factors for aggressive behavior, which resulted in 40 top-ranked and highly interconnected genes. The results of our cross-species and integrated approach provide insights into the genetic etiology of aggression.

Correction to: Psychiatric genetics and the structure of psychopathology.

In the Table 1 legend, the reference numbers and symbols were not correctly presented in the footnotes. The corrected footnotes are presented below.

Effort valuation and psychopathology in children and adults.

BACKGROUND: The Research Domain Criteria initiative was launched by the US National Institute of Mental Health to establish a multi-level framework for understanding psychological constructs relevant to human psychiatric disorders, and identified 'effort valuation/willingness to work' as a clinically useful construct worthy of further study. This construct encompasses the processes by which the cost(s) of obtaining an outcome are calculated, and the tendency to overcome response costs to obtain a reinforcer. The current study aims to examine effort valuation as a correlate of psychopathology in children and adults, and the moderating effects of sex on this relationship. METHODS: Participants were 1215 children aged 6-12 and their parents (n = 1044). All participants completed the Effort Expenditure for Rewards Task as a measure of effort expenditure. Child psychopathology was measured via the Child Behavior Checklist, while adult psychopathology was measured via the Adult Self Report. Additionally, the Social Adjustment Inventory for Children and Adolescents and Injury Behavior Checklist were used to examine child social impairments/problem behaviors. RESULTS: In children, significant interactions between reward sensitivity and sex were observed in association with anxiety and thought problems, specifically at low reward sensitivity levels. In adults, main effects of effort expenditure were seen in drug and alcohol abuse, where higher effort was associated with higher degrees of abuse. CONCLUSIONS: These results establish effort valuation as a relevant psychological construct for understanding psychopathology, but with different profiles of associated psychopathology across sex in children and adults.

Solving for X: Evidence for sex-specific autism biomarkers across multiple transcriptomic studies.

Autism spectrum disorder (ASD) is a markedly heterogeneous condition with a varied phenotypic presentation. Its high concordance among siblings, as well as its clear association with specific genetic disorders, both point to a strong genetic etiology. However, the molecular basis of ASD is still poorly understood, although recent studies point to the existence of sex-specific ASD pathophysiologies and biomarkers. Despite this, little is known about how exactly sex influences the gene expression signatures of ASD probands. In an effort to identify sex-dependent biomarkers and characterize their function, we present an analysis of a single paired-end postmortem brain RNA-Seq data set and a meta-analysis of six blood-based microarray data sets. Here, we identify several genes with sex-dependent dysregulation, and many more with sex-independent dysregulation. Moreover, through pathway analysis, we find that these sex-independent biomarkers have substantially different biological roles than the sex-dependent biomarkers, and that some of these pathways are ubiquitously dysregulated in both postmortem brain and blood. We conclude by synthesizing the discovered biomarker profiles with the extant literature, by highlighting the advantage of studying sex-specific dysregulation directly, and by making a call for new transcriptomic data that comprise large female cohorts.

Genetic correlations among psychiatric and immune-related phenotypes based on genome-wide association data.

Individuals with psychiatric disorders have elevated rates of autoimmune comorbidity and altered immune signaling. It is unclear whether these altered immunological states have a shared genetic basis with those psychiatric disorders. The present study sought to use existing summary-level data from previous genome-wide association studies to determine if commonly varying single nucleotide polymorphisms are shared between psychiatric and immune-related phenotypes. We estimated heritability and examined pair-wise genetic correlations using the linkage disequilibrium score regression (LDSC) and heritability estimation from summary statistics methods. Using LDSC, we observed significant genetic correlations between immune-related disorders and several psychiatric disorders, including anorexia nervosa, attention deficit-hyperactivity disorder, bipolar disorder, major depression, obsessive compulsive disorder, schizophrenia, smoking behavior, and Tourette syndrome. Loci significantly mediating genetic correlations were identified for schizophrenia when analytically paired with Crohn's disease, primary biliary cirrhosis, systemic lupus erythematosus, and ulcerative colitis. We report significantly correlated loci and highlight those containing genome-wide associations and candidate genes for respective disorders. We also used the LDSC method to characterize genetic correlations among the immune-related phenotypes. We discuss our findings in the context of relevant genetic and epidemiological literature, as well as the limitations and caveats of the study.

Interaction Between the µ-Opioid Receptor Gene and the Number of Heavy-Drinking Peers on Alcohol Use.

BACKGROUND: The presence of heavy-drinking peers may trigger genetic vulnerabilities to alcohol use. Limited correlational findings, albeit mixed as a function of age, suggest that carriers of a µ-opioid receptor (OPRM1) G allele may be more vulnerable than noncarriers to alcohol-promoting perceived peer environments. However, research has not yet examined such genetic susceptibility to actual (rather than perceived) peer environments through an experimental, ad libitum alcohol administration design. This study examined whether OPRM1 modulates the effects of heavy-drinking group size on alcohol consumption and explored potential mediators of such OPRM1-based differences. METHODS: Caucasian young adult moderate to heavy drinkers (N = 116; mean age = 22 years [SD = 2.21], 49% female) were randomly assigned to consume alcohol in the presence of none, 1, or 3 heavy-drinking peer confederates. RESULTS: Results showed no significant moderating effects of OPRM1 in the relationship between the number (or presence) of heavy-drinking peers and voluntary alcohol consumption (partial η2  = 0.01). This result remained the same after controlling for sex, age, and typical drinking quantity as well as their 2-way interactions with OPRM1 and social drinking condition. In addition, OPRM1 did not moderate the peer influence on any proposed mediating variables, including craving for alcohol and subjective responses to alcohol. CONCLUSIONS: Findings suggest no OPRM1-based susceptibility to the number of heavy-drinking peers, adding to the existing mixed findings from correlational studies. Future research on OPRM1-related susceptibility to alcohol-promoting peer environments through meta-analytic synthesis and both experimental and prospective, multiwave designs is needed to resolve these mixed findings.

The interaction between the dopamine receptor D4 (DRD4) variable number tandem repeat polymorphism and perceived peer drinking norms in adolescent alcohol use and misuse.

Peer drinking norms are arguably one of the strongest correlates of adolescent drinking. Prospective studies indicate that adolescents tend to select peers based on drinking (peer selection) and their peers' drinking is associated with changes in adolescent drinking over time (peer socialization). The present study investigated whether the peer selection and socialization processes in adolescent drinking differed as a function of the dopamine receptor D4 (DRD4) variable number tandem repeat genotype in two independent prospective data sets. The first sample was 174 high school students drawn from a two-wave 6-month prospective study. The second sample was 237 college students drawn from a three-wave annual prospective study. Multigroup cross-lagged panel analyses of the high school student sample indicated stronger socialization via peer drinking norms among carriers, whereas analyses of the college student sample indicated stronger drinking-based peer selection in the junior year among carriers, compared to noncarriers. Although replication and meta-analytic synthesis are needed, these findings suggest that in part genetically determined peer selection (carriers of the DRD4 seven-repeat allele tend to associate with peers who have more favorable attitudes toward drinking and greater alcohol use) and peer socialization (carriers' subsequent drinking behaviors are more strongly associated with their peer drinking norms) may differ across adolescent developmental stages.

Optimizing the chances of success in the search for epigenetic biomarkers: Embracing genetic variation.

The emphasis on clinical translation in biomedical research continues to grow. This focus has been particularly notable in those investigators using epigenetic approaches to decipher the biology of complex behavioral disorders. As a result of these efforts, reproducible findings for several disorders, such as smoking, have been generated, giving rise to hopes that biomarkers for other behavioral illnesses would be forthcoming. Unfortunately, that biomedical cornucopia has not yet materialized. In this editorial, we review progress to date and discuss barriers to generating epigenetic biomarkers for complex behavioral disorders. We highlight the need to incorporate information on genetic variation and develop more powerful bioinformatics tools in order to optimize the likelihood of success. We emphasize that searches should focus on clearly defined, readily distinguishable behavioral constructs and suggest that some well-intentioned methods, such as correction for cellular heterogeneity, may actually impede the identification of clinically relevant biomarkers in peripheral blood. Finally, we describe how the understanding created by the development of these biomarkers may lead to more valid animal models of neuropsychiatric illness. We conclude that the prospects for epigenetic biomarkers for complex disorders are bright, but emphasize that the journey to the clinical implementation of these findings will be a slow, iterative process.

Blood transcriptomic comparison of individuals with and without autism spectrum disorder: A combined-samples mega-analysis.

Blood-based microarray studies comparing individuals affected with autism spectrum disorder (ASD) and typically developing individuals help characterize differences in circulating immune cell functions and offer potential biomarker signal. We sought to combine the subject-level data from previously published studies by mega-analysis to increase the statistical power. We identified studies that compared ex vivo blood or lymphocytes from ASD-affected individuals and unrelated comparison subjects using Affymetrix or Illumina array platforms. Raw microarray data and clinical meta-data were obtained from seven studies, totaling 626 affected and 447 comparison subjects. Microarray data were processed using uniform methods. Covariate-controlled mixed-effect linear models were used to identify gene transcripts and co-expression network modules that were significantly associated with diagnostic status. Permutation-based gene-set analysis was used to identify functionally related sets of genes that were over- and under-expressed among ASD samples. Our results were consistent with diminished interferon-, EGF-, PDGF-, PI3K-AKT-mTOR-, and RAS-MAPK-signaling cascades, and increased ribosomal translation and NK-cell related activity in ASD. We explored evidence for sex-differences in the ASD-related transcriptomic signature. We also demonstrated that machine-learning classifiers using blood transcriptome data perform with moderate accuracy when data are combined across studies. Comparing our results with those from blood-based studies of protein biomarkers (e.g., cytokines and trophic factors), we propose that ASD may feature decoupling between certain circulating signaling proteins (higher in ASD samples) and the transcriptional cascades which they typically elicit within circulating immune cells (lower in ASD samples). These findings provide insight into ASD-related transcriptional differences in circulating immune cells. © 2016 Wiley Periodicals, Inc.