What

Biomarker discovery and validation across Alto’s data: EEG, ERP, wearable, speech, and cognitive signals modelled across trial cohorts and legacy datasets, using signal processing, statistics, and machine learning to connect a physiological measurement to a clinical outcome.

Resting-state EEG — recorded with no task at all — is the hardest part of it, because it is the cheapest signal to collect and the least constrained to interpret. One resting feature, gamma-band sample entropy, became the patient-selection biomarker for ALTO-300. A 2025 SOBP poster showed that dopamine depletion increases it in humans (d=0.94, p=0.006) and that the same shift appears in mice, linking the compound’s dopaminergic mechanism to the marker used to select patients for it.

Why it mattered

Precision psychiatry only works if a measurement taken before treatment predicts who responds to it. The dosing, the endpoints, and the statistics of a trial all assume that selection step is real. Validating a marker mechanistically, and not only correlationally, is what makes it defensible enough to enrol patients on.

My role

Joined before Alto’s clinical trials were running, into a codebase still being built: research on historical and legacy datasets, patient-stratification modelling, and resting-state EEG preprocessing pipelines built from scratch — infrastructure, Python package development, feature development, literature review, and marker validation. First author on the SOBP 2025 dopamine-depletion poster.

Outcome

EEG biomarkers from this work were used in Phase 2 trials for both major depressive disorder and schizophrenia.