Mobile Menu

Machine learning identifies epigenetic markers of Schizophrenia

We know from twin studies that Schizophrenia carries a 80% heritability risk, but haven’t yet managed to deduce exact genetic risk factors. Epigenetic markers can account for heritability that is not captured by standard genetic association methods, such as polygenic risk scores or twin studies and are being increasingly explored in multifactorial diseases.

Compared to other studies that focussed on a single gene or locus, this study used whole-blood samples to get an overview of the disease. Using machine learning trained on DNA methylation profiles of healthy controls, a polymethylation score (PMS) was generated. This PMS was then applied to 2230 whole-blood samples from 6 global independent cohorts – 36% of which had schizophrenia, and correctly differentiated those with this disorder from control samples.

This DNA methylation signature was then found to be associated with an intermediate phenotype of schizophrenia – those not fully symptomatic – in healthy controls. This linked the epigenetic profile to negative connectivity in the dorsolateral prefrontal cortex hippocampus in relation to working memory. This intermediate phenotype was found in those who are healthy first-degree relatives of schizophrenics, suggesting a potential pre-disposition or risk factor. This could be used to identify those in high-risk groups and go on to explore how this epigenetic mark interacts with other biomarkers of schizophrenia.

By using whole-blood samples and machine learning, this paves the way for PMS to be used in other multifactorial diseases to identify novel targets for drug development or personalised therapeutics.

For those interested in data driven drug discovery/development, join us in Basel later this year at D4 Europe. Find out more about this event and register before the super early bird ends on Monday 24th February

More on these topics

BioData / Epigenetics / Machine Learning