From DNA to drug discovery AI models for cardiovascular precision medicine

Cardiovascular diseases remain the world's leading cause of death-yet the molecular mechanisms linking genetic variation to clinical outcomes are still poorly mapped. Recent advances in artificial intelligence (AI), especially foundation models trained on genomic, structural, and phenotypic data, offer a new opportunity to bridge this gap. This review highlights how AI is reshaping cardiovascular biology through five key domains. First, sequence-to-expression models such as Enformer and Evo2 can predict how noncoding variants affect gene regulation. Second, structure-aware models such as AlphaFold3 and ESM-2 translate sequence into 3D structure, enabling rational design of disease-relevant proteins. Third, cellular context models including scNET and DeepTalk interpret single-cell and spatial transcriptomics to uncover signaling pathways and cell-cell interactions. Fourth, generative chemistry models (e.g. DeepDTAGen, PMDM) design drug-like molecules tailored to protein pockets. Finally, phenotypic response models like PRnet and TranSiGen predict transcriptional changes upon drug treatment-supporting repurposing and toxicity screening. We outline how these components can be integrated into pipelines that trace a path from genotype to phenotype and therapeutic intervention. Use cases include predicting MYH7 variant effects in hypertrophic cardiomyopathy, mapping 9p21 regulatory regions in coronary artery disease, and screening drug toxicity in pulmonary hypertension. Despite progress, a seamless end-to-end system remains out of reach. Future priorities include high-quality datasets, multimodal interoperability, external validation, and clinical integration. Rather than relying on a single model, success will require carefully orchestrated AI pipelines coupled with wet-lab insight. When used strategically, foundation models can compress the discovery-to-clinic timeline, initiating a new era of precision cardiovascular medicine.