Using Deep Learning to Find Genetic Causes of Mental Health Disorders in an Understudied Population

Summary: A new deep learning algorithm that looks for the burden of genomic variants is 70% accurate at identifying specific mental health disorders within the African-American community.

Source: CHOP

Minority populations have been historically under-represented in existing studies addressing how genetic variations may contribute to a variety of disorders. A new study from researchers at Children’s Hospital of Philadelphia (CHOP) shows that a deep learning model has promising accuracy when helping to diagnose a variety of common mental health disorders in African American patients.

This tool could help distinguish between disorders as well as identify multiple disorders, fostering early intervention with better precision and allowing patients to receive a more personalized approach to their condition.

The study was recently published by the journal Molecular Psychiatry.

Properly diagnosing mental disorders can be challenging, especially for young toddlers who are unable to complete questionnaires or rating scales. This challenge has been particularly acute in understudied minority populations. Past genomic research has found several genomic signals for a variety of mental disorders, with some serving as potential therapeutic drug targets.

Deep learning algorithms have also been used to successfully diagnose complex diseases like attention deficit hyperactivity disorder (ADHD). However, these tools have rarely been applied in large populations of African American patients.

In a unique study, the researchers generated whole genome sequencing data from 4,179 patient blood samples of African American patients, including 1,384 patients who had been diagnosed with at least one mental disorder This study focused on eight common mental disorders, including ADHD, depression, anxiety, autism spectrum disorder, intellectual disabilities, speech/language disorder, delays in developments and oppositional defiant disorder (ODD).

The long-term goal of this work is to learn more about specific risks for developing certain diseases in African American populations and how to potentially improve health outcomes by focusing on more personalized approaches to treatment.

“Most studies focus only on one disease, and minority populations have been very under-represented in existing studies that utilize machine learning to study mental disorders,” said senior author Hakon Hakonarson, MD, Ph.D., Director of the Center for Applied Genomics at CHOP.

“We wanted to test this deep learning model in an African American population to see whether it could accurately differentiate mental disorder patients from healthy controls, and whether we could correctly label the types of disorders, especially in patients with multiple disorders.”

The deep learning algorithm looked for the burden of genomic variants in coding and non-coding regions of the genome. The model demonstrated over 70% accuracy in distinguishing patients with mental disorders from the control group. The deep learning algorithm was equally effective in diagnosing patients with multiple disorders, with the model providing exact diagnostic matches in approximately 10% of cases.

The model also successfully identified multiple genomic regions that were highly enriched for mental disorders, meaning they were more likely to be involved in the development of these medical disorders. The biological pathways involved included ones associated with immune responses, antigen and nucleic acid binding, a chemokine signaling pathway, and guanine nucleotide-binding protein receptors.

However, the researchers also found that variants in regions that did not code for proteins seemed to be implicated in these disorders at higher frequency, which means they may serve as alternative markers.

“By identifying genetic variants and associated pathways, future research aimed at characterizing their function may provide mechanistic insight as to how these disorders develop,” Hakonarson said.

Abstract

Application of deep learning algorithm on whole genome sequencing data uncovers structural variants associated with multiple mental disorders in African American patients

Mental disorders present a global health concern, while the diagnosis of mental disorders can be challenging. The diagnosis is even harder for patients who have more than one type of mental disorder, especially for young toddlers who are not able to complete questionnaires or standardized rating scales for diagnosis. In the past decade, multiple genomic association signals have been reported for mental disorders, some of which present attractive drug targets.

Concurrently, machine learning algorithms, especially deep learning algorithms, have been successful in the diagnosis and/or labeling of complex diseases, such as attention deficit hyperactivity disorder (ADHD) or cancer. In this study, we focused on eight common mental disorders, including ADHD, depression, anxiety, autism, intellectual disabilities, speech/language disorder, delays in developments, and oppositional defiant disorder in the ethnic minority of African Americans.

Blood-derived whole genome sequencing data from 4179 individuals were generated, including 1384 patients with the diagnosis of at least one mental disorder. The burden of genomic variants in coding/non-coding regions was applied as feature vectors in the deep learning algorithm. Our model showed ~65% accuracy in differentiating patients from controls. Ability to label patients with multiple disorders was similarly successful, with a hamming loss score less than 0.3, while exact diagnostic matches are around 10%. Genes in genomic regions with the highest weights showed enrichment of biological pathways involved in immune responses, antigen/nucleic acid binding, chemokine signaling pathway, and G-protein receptor activities.

A noticeable fact is that variants in non-coding regions (e.g., ncRNA, intronic, and intergenic) performed equally well as variants in coding regions; however, unlike coding region variants, variants in non-coding regions do not express genomic hotspots whereas they carry much more narrow standard deviations, indicating they probably serve as alternative markers.