Using biomarkers and machine learning to predict antidepressant resistance

Around half of patients with depression don’t improve after taking antidepressants. Clinicians need to make informed decisions on the appropriate therapy for their patients. It would be best if these decisions were tailored to a patient’s particular characteristics, such as their clinical features like age and body mass index, social background and their genes. 

However, the evidence so far on the link between a patient’s characteristics and how they respond to treatment isn’t strong enough for clinicians to use for decision making. We want to develop a biomarker for response to antidepressant treatment, which would better inform clinical practice.  

DNA methylation (DNAm) is a chemical process regulating cellular activity that could be helpful in identifying such biomarkers. We will explore whether DNAm can help predict whether a patient will respond to antidepressants at the point of them being diagnosed.  

DNAm is a promising process in identifying biomarkers as it’s influenced by both genetic and environmental factors, while being stable and easily measured in blood. So far, some studies have identified DNAm as being associated with resistance to anti-depressants. But these studies have been either too small or limited to a small number of sites’ DNAm.  

These studies suggest it might be possible to use DNAm markers to predict if a patient will benefit from antidepressants. Applying machine learning to sociodemographic, clinical and genotype data has made finding markers to predict antidepressant resistance more accurate and replicable. Incorporating DNAm data into these machine learning models should improve these predictions further.  

The Bristol BRC funded GENDEP (Genome-based therapeutic drugs for depression) study collected genome data from 200 participants. GENDEP participants were randomized to a 12-week course of Escitalopram, a type of antidepressant called a selective serotonin-reuptake inhibitor, and blood DNA was collected before treatment.  

We will use the GENDEP data to investigate how well DNAm can predict response to antidepressants. We also aim to develop a corresponding robust DNAm biomarker using machine learning.  

We will use data from the cohort studies Children of the 90s and Generation Scotland as independent test sets, to verify the performance of GENDEP trained DNAm predictors.