Machine learning and multi-omics technologies for precision cardiovascular medicine: advancing diagnosis, risk prediction, and therapeutic guidance

Figure 1. Framework of precision cardiovascular medicine driven by multi-omics and ML. This figure systematically illustrates the workflow and applications of multi-omics and ML in precision cardiovascular medicine research. (1) Data preparation. Blood samples, medical imaging data, and clinical information are collected via well-designed discovery and validation cohorts. Multi-omics technologies, including genomics, transcriptomics, proteomics, and metabolomics, are employed, and raw medical imaging data are processed according to standardized procedures to construct analytical datasets. (2) Machine learning. Appropriate machine learning algorithms are selected for subsequent analysis. Given the feature characteristics and dimensionality of omics data, conventional ML approaches (e.g., linear models such as LR and LASSO, and tree-based models such as RF and XGBoost) and DL frameworks are used to construct models. Post-hoc interpretability tools, such as SHAP, are further integrated to enhance model clarity and transparency. (3) Model building & validation. Established ML models can be implemented in the clinical practices of CVDs, comprising diagnosis assistance (diagnosis confirmation,…

Figure 2. Clinical applications of multi-omics and ML in the diagnosis of CVDs. This figure illustrates the pivotal value of multi-omics and ML in CVD diagnosis in four aspects: confirmatory diagnosis, early diagnosis, differential diagnosis, and severity assessment. (A) Diagnosis Confirmation. Traditional diagnostic workflows for certain CVDs largely rely on invasive examinations, whereas non-invasive modalities are limited by semi-quantitative evaluation and the disconnection between functional and anatomical evidence. ML can effectively mitigate these limitations and facilitate comprehensive clinical management of cardiovascular comorbidities. (B) Early diagnosis. ML enables the detection of molecular perturbations at the subclinical stage via molecular omics profiling and captures subtle textural alterations in imaging data to identify minimal early structural and functional abnormalities. This strategy allows preclinical recognition of asymptomatic CVDs and provides a critical time window for early therapeutic intervention (29, 30). (C) Differential diagnosis. ML can distinguish different subtypes of the same disease, as well as distinct diseases with overlapping imaging phenotypes or…

Figure 3. Applications of multi-omics and ML in CVD risk prediction and therapy response assessment. This figure illustrates the clinical translation value of multi-omics and ML from two angles: risk prediction and therapeutic response assessment. Left panel: Risk prediction (A) Primary prevention. This analytical framework provides incremental prognostic information beyond conventional clinical risk scores such as SCORE2, enabling reliable identification of high-risk individuals. It also offers a feasible methodological approach to opportunistic screening for CVDs (51, 54). (B) Secondary prevention. Secondary prevention aims to assess residual risk among patients with established CVD. Multi-omics, integrated with ML, enables characterization of pathophysiological processes closely linked to disease progression, thereby providing more comprehensive evidence for refined risk stratification and clinical management (55–57). (C) Short-term risk prediction. ML models based on metabolomic signatures capture acute pathophysiological activity and provide early warning of recurrent adverse events. This is particularly critical for high short-term-risk conditions such as ACS (58). (D) Specific…