BioSync Uses Transformer Fusion to Build a Multimodal Physiological Biomarker

Summary

BioSync is a Transformer-based architecture that combines cardiac, neural, behavioral, and speech measurements from wearable and mobile devices into the continuous BioSync Index (BSI), defined under the BEST framework. It uses multi-head self-attention over modality tokens together with a linear branch whose hypothesis class includes standard feature concatenation. The authors evaluated it on two literature-informed synthetic cohorts: a cognitive-decline cohort using HRV, EEG, actigraphy, and speech, and a metabolic-autonomic cohort structured around the public AI-READI wearable schema. On the cognitive cohort, BioSync achieved an AUC of 0.928 versus 0.926 for concatenation; on the metabolic cohort, it reached 0.764 accuracy and 0.766 F1, compared with 0.756 and 0.758. The BSI correlated with latent severity at r=0.91 and r=0.68 in the two cohorts. A pure-attention ablation reached 0.911 AUC, indicating that the gain was associated with the combined wide-and-deep architecture. With matched modality-dropout training, BioSync outperformed concatenation at five of six cognitive corruption rates and at the highest metabolic corruption rate. Its cognitive AUC was higher than five published reference values, but dataset and task differences prevent a controlled benchmark claim. The authors state that validation on real cohorts is still required.