DISENTANGLING TIGHTLY LINKED RISK AND RESILIENCE GENES IN ADHD

Research on the genetic and biological capacity to withstand risk for disease necessarily lags behind the discovery of reliable risk factors. Yet, as the reliability of ADHD-risk discoveries increases, so does the feasibility of ADHD-resilience research. Knowing how some people avoid disease despite being at high risk may shed light on novel targets for treatment, intervention, and prevention that could not be found by studying affected individuals alone. However, resilience genes are difficult to detect because protective resilience variants may be masked by risk variants. We developed an adversarial learning (AL) approach that explicitly separates resilience from risk by training a machine learning model to learn patterns that discriminate high-risk controls from risk-matched cases while actively “unlearning” patterns present in low-risk groups. We applied this approach to simulated data and an ADHD dataset with over 50,000 participants. We identified markers of resilience with a feature-importance-based approach that prioritized specificity, generated polygenic resilience scores, and tested for an interaction between polygenic risk and resilience scores. In simulations, our approach had high specificity (0.99) and sensitivity (0.95) for identifying resilience markers, significantly outperforming traditional approaches. Applied to ADHD data, we identified 45 resilience markers and found a significant interaction (p = 2.9 × 10-11) between polygenic risk and resilience, such that high resilience scores attenuated the effect of polygenic risk. Our findings support the utility of resilience scores in modifying risk predictions, particularly for high-risk groups. Expanding this method could aid in understanding resilience mechanisms, potentially improving diagnosis, prevention, and treatment strategies for ADHD and other psychiatric disorders.