Cardiovascular diseases (CVDs) persist as a predominant cause of mortality globally, underscoring the urgent need for innovative diagnostic and therapeutic strategies. Multimodal artificial intelligence (AI) and machine learning (ML) offer transformative potential in this domain, particularly through the integration of multi-omics data. This integration allows researchers to identify novel biomarkers and enhance disease prediction, steering the field towards precision medicine. This article explores the methodology of using multimodal AI/ML to discover novel biomarkers and predict disease using the multi-omics profiles of patients with cardiovascular diseases.
The integration of multi-omics data begins with meticulous pre-processing, a critical phase in managing complex biological datasets. A study involving 71 participants, including individuals with CVDs and healthy controls, utilised RNA sequencing (RNA-seq) to extract transcriptomic data from peripheral blood mononuclear cells (PBMCs). This data collection adhered to stringent ethical standards. Initial datasets underwent rigorous filtering to eliminate transcripts with low expression levels and those not consistently expressed across all participants. The refined dataset, formatted into a clinically integrated transcriptomics and genomics dataset (CIGT), formed the foundational basis for subsequent analyses.
To address the issue of missing data, researchers employed the k-nearest neighbours (k-NN) imputation technique. This method simulated missing values and optimised parameters, thereby minimising noise and enhancing the integrity of the data. The imputed dataset then underwent normalisation using DESeq2’s median-of-ratios method, ensuring cross-sample comparability and the preservation of biological signals. This meticulous pre-processing was essential for accurate downstream analyses, laying the groundwork for the next phase of the study.
The subsequent stage focused on transcriptomic and gene expression analysis using DESeq2. To mitigate potential confounding factors, the cohort was stratified into subcohorts based on demographic features such as sex, race, and age. Differential expression analysis was conducted independently within each subcohort, revealing differentially expressed genes (DEGs) with significant p-values. These results were consolidated to identify DEGs crucial for CVD prediction. To refine this selection further, researchers employed a minimum redundancy – maximum relevance (MRMR) approach, which minimised redundant information and highlighted biomarkers most relevant to distinguishing between patients and controls. Gene Set Enrichment Analysis (GSEA) provided additional insights into the biological processes and disease implications associated with these biomarkers.
The integration of whole-genome sequencing (WGS) data added further depth to the analysis. Single nucleotide polymorphisms (SNPs) were processed and annotated using tools like the Ensembl Variant Effect Predictor (VEP) and Combined Annotation Dependent Depletion (CADD). By concentrating on SNPs associated with MRMR-selected DEGs, researchers were able to focus on genomic regions implicated in CVD. The deleteriousness of SNPs was assessed using CADD scores, identifying rare variants with potential pathogenic impact. This comprehensive approach reduced confounding data, enhancing the likelihood of identifying significant genetic contributors to CVD, and provided a holistic view of the genetic landscape in CVD patients.
The final stage was the integration of selected DEGs and SNPs into an AI/ML-ready dataset. This dataset underwent rigorous analysis using machine learning classifiers such as Random Forest, eXtreme Gradient Boosting (XGBoost), and Logistic Regression. Bayesian optimisation was employed to fine-tune hyperparameters, maximising the performance of each classifier. The classifiers were evaluated based on metrics like accuracy, sensitivity, and specificity. SHapley Additive exPlanations (SHAP) scores offered insights into the importance and directionality of each feature in predicting CVD. By combining SHAP profiles with prediction probabilities, researchers identified key biomarkers contributing to CVD prediction.
The integration of multimodal AI/ML with multi-omics data represents an exciting frontier in cardiovascular disease research. By uncovering novel biomarkers and enhancing disease prediction, this approach heralds a new era in personalised medicine. As technology continues to advance, the potential to transform CVD diagnosis and treatment becomes increasingly tangible, offering hope for improved patient outcomes and a deeper understanding of this complex disease. The convergence of these cutting-edge methodologies promises not only to revolutionise the field of CVD diagnostics but also to significantly improve the quality of life for patients worldwide.
Be the first to comment