Scientists use machine learning to crack it

SAN FRANCISCO—May 24, 2024—In a scientific feat that expands knowledge about genetic changes that shape brain development and cause mental disorders, a team of researchers has combined high-throughput experiments and machine learning to analyze more than 100,000 sequences of human brain cells and identify more than 150 mutations that may cause disease.

The study, by scientists from the Gladstone Institutes and the University of California, San Francisco (UCSF), establishes a comprehensive catalogue of gene sequences involved in brain development and opens the door to new diagnostics and treatments for neurological disorders such as schizophrenia and autism spectrum disorder. The findings are published in the journal Neuroscience. Science.

“We collected a huge amount of data from the sequence of non-coding regions of DNA that are already suspected to play major roles in brain development and disease,” said senior research scientist Katie Pollard, PhD, who also serves as director of the Gladstone Institute for Data Science and Biotechnology. “We were able to functionally test over 100,000 of them to see if they affect gene activity, pinpointing sequence changes that may alter activity in disease.”

Pollard co-led the extensive study with Nadav Ahitov, PhD, professor in the UCSF Department of Bioengineering and Therapeutic Sciences and director of the UCSF Institute of Human Genetics. Much of the experimental work on the brain tissue was led by Tomasz Nowakowski, PhD, associate professor of neurosurgery in the UCSF School of Medicine.

In total, the team found 164 mutations associated with psychiatric disorders and 46,802 sequences with enhancer activity in developing neurons, meaning they control the function of certain genes.

These “enhancers” could potentially be used to treat psychiatric disorders in which one copy of a gene is not fully functional, Ahitubu said. “Hundreds of diseases result from one gene not functioning properly, and we may be able to harness these enhancers to make that gene work better.”

Organoids and machine learning take center stage

Beyond identifying enhancers and disease-associated sequences, this study has implications in two other important areas.

First, the scientists repeated some of their experiments using brain organoids made from human stem cells and found that the organoids were effective surrogates for real brains: Remarkably, most of the genetic mutations detected in human brain tissue were reproduced in the brain organoids.

“Our organoids are very similar to the human brain,” Ahitov says, “and as we expand our work to test for an array of other neurodevelopmental diseases, we are finding that organoids are an excellent model for understanding gene regulatory activity.”

Second, by feeding large amounts of DNA sequence data and gene regulatory activity into a machine learning model, the team was able to train a computer to accurately predict the activity of specific sequences. This type of program enables “in-silico” experiments, allowing researchers to predict experimental outcomes before conducting them in the lab. This strategy allows scientists to make discoveries more quickly using fewer resources, especially when large amounts of biological data are involved.

Dr. Shawn Whalen, a senior research scientist in Gladstone University's Pollard Laboratory and co-first author of the study, said the team tested their machine learning model using sequences retained from model training to see if it could predict outcomes already collected about gene expression activity.

“The model had never seen this data before, but it was able to make a very accurate prediction, indicating that it had learned a general principle about how genes are influenced by non-coding regions of DNA in developing brain cells,” Whalen said. “You can imagine that this will open up many new possibilities in research, such as predicting how combinations of mutations will work together.”

A new chapter in brain discovery

This study was completed as part of the PsychENCODE Consortium, which brings together multidisciplinary teams to generate large-scale gene expression and regulatory data from the human brain across several major psychiatric disorders and stages of brain development.

Through publishing multiple research papers, the consortium aims to shed light on poorly understood mental illnesses, from autism to bipolar disorder, and ultimately introduce new treatments.

“Our study contributes to this growing body of knowledge by demonstrating the utility of using human cells, organoids, functional screening methods, and deep learning to investigate regulatory elements and mutations involved in human brain development,” said Chengyu Deng, PhD, a postdoctoral researcher at UCSF and co-first author of the study.

About the Research

The study, “Massively parallel characterization of regulatory elements in the developing human cerebral cortex,” appears in the May 24, 2024 issue of the journal Neuroscience. Science. Authors include Chengyu Deng, Sean Whalen, Marilyn Steyert, Ryan Ziffra, Pawel Przytycki, Fumitaka Inoue, Daniela Pereira, Davide Capauto, Scott Norton, Flora Vaccarino, the PsychENCODE Consortium, Alex Pollen, Tomasz Nowakowski, Nadav Ahituv, and Katherine Pollard.

This research was funded in part by the National Institute of Mental Health, the New York Stem Cell Foundation, the National Human Genome Research Institute, and the Office of the Coordinator for Advancement of Higher Education and Human Resources. Data generated are part of the PsychENCODE consortium.

About Gladstone Institutes

The Gladstone Institutes are independent, nonprofit life science research institutes using visionary science and technology to defeat disease. Founded in 1979, we are located in the biomedical and technological innovation hub of San Francisco's Mission Bay district. Gladstone Institutes has transformed the way science is done, creating a research model that funds big ideas and attracts the best and brightest minds.