AI-ENHANCED DATA SCIENCE ARCHITECTURES: A REVIEW OF AUTOMATED LEARNING, ANALYTICS, AND DECISION SUPPORT

Abstract

The explosion of genomic, transcriptomic, proteomic, metabolomic and other multi-omics data has resulted in many opportunities to advance understanding of complex biological processes and a significant computational and analytical challenge. To overcome such challenges, the use of AI powered data science architectures is growing, enabling automated learning, scalable data integration, predictive modelling, and decision-support systems. This review explores the place of AI-enabled architectures in genomic and molecular studies focusing on data collection and pre-processing, computational infrastructure, machine learning, deep learning, automated learning and biological interpretation. AI-based multi-omics integration, genomic prediction, biomarker discovery, cancer genomics, pharmacogenomics, molecular diagnostics and precision medicine are discussed. It also features discussions about emerging techniques, such as transformers, graph neural networks, generative AI, multimodal learning, biological foundation models, and federated learning, and how they can enhance molecular representation and clinical translation. Although significant progress has been made, significant limitations remain around model interpretability, model reproducibility, population bias, data heterogeneity, genomic privacy and external validation. There should thus be a focus in future genomic systems using AI for transparency in modelling, standardisation of analysis workflow, biologically meaningful interpretation and clinically responsible deployment. In sum, AI-driven data science architectures hold significant potential to revolutionize the way complex molecular data is analysed to yield meaningful insights into biology and precision-driven decision-making.