Google's AI Blog introduces the Biomarker Discovery Framework, a multi-agent system that structures candidate biomarker prioritization from wearable sensor data as an iterative research loop under human supervision. By combining hypothesis generation, parallel statistical analysis, model training, adversarial validation, and literature-grounded reasoning, it accelerates discovery while maintaining strict statistical rigor.

An Orchestrator agent decomposes natural-language research directives into execution plans, guiding specialized agents through six phases: data understanding, hypothesis grounding, an iterative discovery loop, adversarial validation, deep research and assessment, and report writing. Shared memory, a structured fact sheet, and common tools preserve traceability across the workflow.

Across three cohorts totaling 9,279 participant-observations, the framework identified 41 candidate digital biomarkers for mental health and 25 for metabolic outcomes. In DWB, sleep-duration variability was associated with PHQ-8 severity (ρ = 0.252), framed as a literature-grounded circadian-instability hypothesis. Framework-derived features improved downstream prediction when combined with demographic variables (ΔR² = 0.040 for depression).

In a blinded evaluation, 15 experts scored its reports against three contemporary AI systems. The framework received the highest mean scores across all seven quality dimensions and was the only system recommended for Accept or Minor Revision. Reviewers estimated they would retain 56.9% of its manuscript content on average, versus 18.8%–30.4% for baselines.