Applying multimodal biological foundation models across therapeutics and patient care

AWS is advancing healthcare AI with multimodal biological foundation models (BioFMs) that integrate fragmented data types—genomics, medical images, clinical notes, electronic health records, and drug discovery data—into unified systems for therapeutics and patient care. Traditional approaches analyze these data streams separately, missing critical insights across modalities. Multimodal BioFMs, like Latent Labs' Latent-X1/X2 for protein design, Arc Institute's Evo 2 for DNA/RNA modeling, and John Snow Lab's Medical VLM-24B for diagnostics, simultaneously process text, images, and molecular data to deliver richer insights. These models achieve 4-7% higher diagnostic accuracy (AUC) for conditions like Alzheimer's and brain cancer compared to unimodal baselines, according to Sun et al. 2024. AWS provides scalable compute and partner tools to build and deploy these models across the drug development lifecycle, from research to clinical validation.

Real-world applications span drug discovery, clinical development, and personalized medicine. For example, Insilico Medicine's Nach01 integrates chemical intelligence and 3D molecular data to accelerate drug candidate identification, while Harvard and AstraZeneca's MADRIGAL predicts drug combination outcomes and adverse interactions using structural and transcriptomic data. GE Healthcare's 3D MRI foundation model enables image retrieval and segmentation tasks. AWS's unified environment supports organizations like Bioptimus and Latent Labs, enabling faster, more confident decision-making. By breaking down data silos, multimodal BioFMs promise to revolutionize how healthcare professionals diagnose diseases, prescribe treatments, and predict outcomes, ultimately improving patient care and reducing costs.