Longitudinal alignments and syntheses of multimodal clinical data for personalized medicine with the PULSE framework.

PULSE accurately reconstructs metabolomic and proteomic profiles from sparse blood tests, outperforming benchmarks and enabling disease prediction with AUROCs up to 0.83.

• Why it matters: Understanding physiological dynamics through longitudinal, multimodal data is crucial for personalized medicine, yet current methods struggle to handle the temporal and diverse nature of patient data.
• What they did: The PULSE framework employs self-supervised learning to encode past patient states and reconstruct full profiles from unpaired measurements, integrating various data types like images and health records.
• The result: This approach captures continuous disease physiology, improves multimodal alignment, and enables robust feature extraction, facilitating accurate disease prediction and advancing personalized healthcare.