Validating of Machine Learning-Based EEG Treatment Biomarkers in Depression

SUMMARY/ABSTRACT The overarching aim of Alto Neuroscience is to advance brain-based biomarkers for psychiatric disorders in order to both optimize treatment pathways and drive the development of novel pharmacological and non- pharmacological interventions. Alto does this by developing and applying sophisticated machine learning computational models to electroencephalography (EEG) data collected at scale in real-world clinical treatment contexts. Specifically, in this direct-to-phase II SBIR proposal we will refine, and then independently validate, two EEG-based candidate biomarkers we have identified for stratifying patients with depression in a manner that both factors biological heterogeneity and informs treatment response. One of our biomarkers was derived in a “top-down” (i.e. supervised) manner by trying to directly predict treatment outcome, while the other biomarker presents a complimentary “bottom-up” (i.e. unsupervised) approach that begins by first identifying the most biologically homogeneous subset of patients and then testing the treatment relevance of the subtyping. Together, these findings represent very robust individual patient-level treatment-relevant EEG biomarkers, and in both cases, help define a critically-important objective approach to prospectively identifying and treating treatment- resistant depressed patients. A successful outcome of the proposed work would yield the first FDA-cleared biomarkers for stratifying psychiatric conditions. It would also provide a basis for targeted development of pharmacological and non-pharmacological interventions based on the EEG biomarkers. Both outcomes hold substantial commercial value and exciting potential for transforming psychiatry.