Personalized medicine in autism requires stratification biomarkers because of substantial biological heterogeneity. However, there is still a lack of any validated stratification biomarkers. The identification of such biomarkers requires direct, sensitive and individualizable measures of variation in key neurobiological systems. Current discovery approaches use a highly constrained set of cognitive/perceptual tasks, coupled with a highly unconstrained set of analytic pipelines. Case-control studies designed to test targeted hypotheses based on the existing literature often fail because group-level results do not necessarily yield meaningful individual-level variation, and the existing literature contains many false positives due in part to analytic flexibility. At the other extreme, purely data-driven attempts to associate every neuroimaging phenotype with behavior require huge sample sizes, cause dataset decay and have not delivered individual-level insights.

Emily Jones and Sarah Lippé are planning to initiate a program of research known as GAIINS (Genetics and Artificial Intelligence for Individualized Neural Stratification). GAIINS aims to capitalize on new developments in AI that will enable the research team to efficiently search for individualized stratification biomarkers across a broad task and analytic space by harnessing the power of large-scale datasets but without exhaustively accessing data. The experimental dataset will consist of electroencephalogram (EEG) recordings from children with autism. EEG is a scalable tool that represents a direct measure of neuronal activity. If successful, this approach is designed to be generalizable across methods and systems.

Jones and Lippé’s approach is designed to prioritize neurobiological sensitivity and individualization in the discovery process and builds-in external validation. First, the research team plans to build a pipeline to compute a broad set of EEG features and map their covariation. Second, they aim to use this pipeline and a new AI-based approach to identify individual-level EEG features that are sensitive to change in a randomized clinical trial of a medication that targets a particular neurotransmitter system and examine generalization to a second trial. Third, they aim to validate the utility of these EEG features as prognostic and/or mechanistic biomarkers at the neurobiological and clinical level in external genetically characterized cohorts, including SPARK. Taken together, GAIINS is expected to harness the power of AI to revolutionize the way that stratification biomarkers are identified for autism.