This story has been updated from a version posted April 14 to clarify the non-clinical nature of the project described below.
NEW YORK – Yale University researcher Thomas Fernandez has teamed up with bioinformatics startup Allelica to develop polygenic risk scores (PRSs) related to certain neuropsychiatric disorders such as Tourette syndrome, obsessive-compulsive disorder (OCD), and autism spectrum disorders (ASD).
Using the company's Discover platform and funded by a grant from the National Institutes of Health, Fernandez, a child psychiatrist and psychiatric geneticist, aims to identify PRS predictive of heightened risk of developing these conditions and to better understand their underlying biology.
Fernandez has studied the role of rare genetic variation in child psychiatric disorders for roughly the past 10 years. In general, the field tends to focus on rare genetic variants, he said, as these will have a higher impact than common variants.
The interaction between rare and common variants, however, may guide researchers to a better understanding of a disorder's underlying biology. "I think the hypothesis would be that both would kind of converge on shared pathways for a disorder," he said. "But it's quite possible that we may see distinct pathways as well."
On the clinical side of things, Fernandez hopes that PRSs might also help clinicians differentiate who is more or less likely to respond to a given intervention.
"As a child psychiatrist," he said, "it is very difficult to predict who's going to respond, for example, to [the] interventions we have. Wouldn't it be great if we could figure out who we need to monitor more closely?"
As an example of a clinical application of polygenic risk, Fernandez cited recent advances in the use of PRS to stratify cardiovascular disease risk, which can help physicians identify people who might benefit more from early interventions such as statin therapy.
Last year, for example, Allelica presented evidence that PRSs could improve risk estimations of coronary artery disease over a patient's low-density lipoprotein levels alone.
However, Renato Polimanti, a geneticist at Yale with a focus on biological psychiatry, who was not involved in this project, cautioned that the science of PRS for neuropsychiatric disorders has not yet shown strong predictive powers at the individual level.
"PRS are an invaluable instrument to study the genetic architecture of psychiatric disorders and their comorbid traits," he said. "However, although PRS for neuropsychiatric disorders can provide highly significant associations at a population level, their predictive power is too low to stratify disease risk at an individual level."
Although Fernandez sees "a lot of clinical value" in PRS, individual risk prediction falls outside the scope of the current project, which seeks to uncover associations between rare variants and polygenic risk, thereby shedding light on the underlying biology of neuropsychiatric disorders.
Much of the work will involve breaking new ground, as Fenandez noted that "the mechanisms of cancer are much better worked out than the mechanisms of psychiatric disease."
Despite the caveat, Polimanti noted that PRS for certain medical outcomes, such as for cardiovascular diseases and some cancers, are approaching clinically useful predictive powers. The issue with neuropsychiatric disorders that may make them more complicated, he said, is that potentially thousands of genetic variants with small individual effects can contribute to a predisposition to psychiatric disorders.
To assist in discovering and validating Fernandez's PRSs, Allelica has donated the use of its Discover tool, a software-as-a-service platform consisting of 11 algorithms designed to help researchers build PRS from genetic data and summary statistics from a genome-wide association study.
Fernandez reached out to Allelica roughly seven months ago regarding a collaboration and the use of the company's tools, which he described as user-friendly for a non-bioinformatician.
One key deciding factor in using the Discover platform, he explained, was that its various algorithms all work with the same file formats.
"Often, different algorithms require differently formatted files," said George Busby, Allelica's cofounder and CSO. "Moreover, Discover handles the running of these algorithms in parallel on high-performance computing clusters/cloud-based services, meaning that there is minimal need for additional bioinformatic work to get to the results."
Discover uses a variety of algorithms to generate PRSs. These include LDpred2, LDpred2-auto, PRS-CS, PRS-CS-auto, PRS-CSx, stacked clumping and thresholding, Clumping and Thresholding, Lassosum, SBayesR, Support Vector Machine, and GWAS significant.
"What this allows you to do is very quickly identify both the methodology [and] the best PRS panel that comes from your data," Busby said.
Allelica is providing its Discover service to Fernandez free of charge and views the collaboration as a way to enable PRS research that will inform later commercial opportunities and clinical care.
"It's really enabling Tom to do his PRS research in a robust and scalable [way], and crucially, in a quicker way than would be possible without this kind of unified piece of software that we're providing for him," Busby said.
Although Allelica's role in the collaboration so far has been to provide access to its technology, the firm expects to help analyze data as it becomes available.
Initially, Fernandez plans to conduct retrospective studies, using genotyping array, GWAS, and whole-genome data from prior studies and public databases, such as the Simons Foundation Powering Autism Research (SPARK) database, which contains over 3,000 genomes and over 31,000 exomes of people with ASD. Fernandez hopes to follow these with prospective studies after the retrospective data has been analyzed.
He noted that relatively little work on PRS has been done in the cases of Tourette syndrome and OCD, making it a fertile area for discovery.
In the case of ASD, Fernandez plans to look in particular at children with aggression issues and stereotyped behaviors, as these both present urgent medical needs and because they present relatively well-defined phenotypes in an otherwise highly heterogeneous disorder.
"We had the hypothesis that if we can subdivide the autism spectrum disorder into phenotypes, then those phenotypes might be a little more genetically homogeneous," he said.
The large datasets that Fernandez will analyze are in some ways ideal for finding polygenic associations and developing testable hypotheses, as they contain large sample sizes with accompanying clinical and phenotypic data.
However, they do suffer from a lack of genetic diversity that perennially plagues the wider field of genomic research because they tend to be heavily biased toward people of European descent, which can limit the portability of any resulting PRS.
Large biobanks can also be affected by participation biases related to the population from which they are drawn.
"This could create an issue in translating PRS across different settings," Polimanti said, "especially when considering psychiatric disorders because participation biases appear to be strongly associated with certain behavioral traits."
Funding agencies and international collaborations are trying to fill these gaps, and diverse prospective studies, such as those Fernandez hopes to be able to conduct in the near future, will be critical to driving the science and utility of PRS forward.
Although PRS alone will never fully predict an individual's disease risk, especially in complex cases like neuropsychiatric disorders, Busby hopes they will contribute to risk predictions.
"I'd like to see a way in which polygenic risk is used as part of the suite of pieces of information and tools that can really help understand who's at high risk and then actually do something about that in the situations where there are ways in which you can mitigate that risk," he said, "or even as a way of just trying to understand why disease happens in some people and not in others."