Development and validation of AI-driven multi-omics language models for cancer genomics: A comprehensive review

The pervasive challenges in cancer management, ranging from accurate early diagnosis to effective personalised therapies and precise patient stratification, represent significant clinical unmet needs. Artificial intelligence (AI) is transforming cancer research by offering unprecedented capabilities in analysing complex genomic datasets. AI has completely transformed omics research by simplifying the integration of multi-omics data, offering more profound insights into cancer heterogeneity, and enhancing predictive models for patient treatment responses. In the past decade, numerous studies have highlighted how AI has revolutionised omics research. AI models are instrumental in enhancing oncology by addressing these unmet needs through improved clinical trial matching, refined risk assessment, and precise treatment selection. They contribute to more personalised and effective cancer care by classifying cancer types and subtypes, identifying biomarkers, predicting drug responses, stratifying patients, and analysing tumour evolution and heterogeneity. This comprehensive review specifically focuses on the development and validation of AI-powered multi-omics language models for cancer genomics. We posit that the integration of diverse omics data types provides synergistic insights beyond single-omics approaches, which are critical for unravelling cancer heterogeneity and addressing the complex challenges within cancer genomics. The review highlights recent advances, current difficulties, and potential paths forward for these integrated AI approaches. We also detail the main elements of these models, such as their architectures, training plans, evaluation techniques, and data preprocessing. All things considered, multi-omics language models powered by AI hold immense promise for deriving biological insights from intricate cancer genomic data and converting them into actionable information for clinical settings.