Leveraging artificial intelligence to integrate genomics, transcriptomics, and proteomics data for enhanced disease prediction

Abstract

The rapid expansion of high‑throughput sequencing and mass‑spectrometry technologies has given rise to vast amounts of genomics, transcriptomics and proteomics data—offering unprecedented insight into biological systems. However, analysing each data type in isolation often fails to capture the cross‑layer complexity of gene regulatory networks, post‑translational modifications and phenotypic manifestations. To address this gap, artificial intelligence (AI) algorithms—particularly supervised machine learning, deep neural networks and multimodal fusion models—are increasingly employed to integrate multiple “‑omics” layers for systems‑level insight. In this paper, we examine the development and deployment of AI‑based pipelines that combine genomic sequence/variant data, RNA expression profiles and proteomic abundance measurements. We outline key methodological steps: data preprocessing and normalization, feature engineering across omics, architecture choices (e.g., autoencoders, graph neural networks, attention‑based fusion), training and validation workflows. A hypothetical benchmarking dataset (n = 500 patients, three omics layers) illustrates how a multimodal fusion model improved disease classification accuracy (AUC ~ 0.92) versus single‑omics models (~ 0.83), and revealed novel cross‑layer biomarkers. We discuss advantages (higher predictive power, ability to discover cross‑layer signatures), as well as challenges: data heterogeneity, missing modality data, interpretability, and generalisability across cohorts. Finally, future perspectives are presented: self‑supervised foundation models across omics, federated learning for privacy‑sensitive data, and explainable AI (XAI) to enhance clinical trust. In conclusion, the integration of genomics, transcriptomics and proteomics via AI holds strong promise for deeper mechanistic insight and precision medicine—but realising that promise depends on methodological rigor, interpretability and equitable representation of diverse populations.