Project Funding Details

Artificial intelligence (AI) is revolutionizing oncology by leveraging deep learning techniques for enhanced image processing in pathology and radiology. The integration of AI with multi-omics data-comprising genomics, transcriptomics, proteomics, and metabolomics-promises a comprehensive understanding of cancer heterogeneity, essential for personalized oncology. However, appropriate modelling of the biological interdependence among omics layers, the handling of missing data and data scarcity remain significant challenges to wider adoption of AI multi-omics models in oncology. The integration of multi-omics data using AI models can yield robust prognostic biomarkers, enhancing the precision of cancer diagnosis and treatment. We hypothesise that i) incorporating cross-omics learnings can overcome challenges related to the biological interdependence among omics layers, ii) data augmentation and improved AI models can help with missing data and iii) the incorporation of biological insights from complementary omics assays provide added knowledge to overcome data scarcity. 1. Address challenges of omics interdependence, missing data and data scarcity in the development of multi-omics AI models in oncology 2. Develop a multi-omics AI framework for the identification of prognostic biomarkers 3. Apply these models to hepatocellular carcinoma (HCC) to identify prognostic biomarkers and improve patient stratification WP1. Data Augmentation: For our existing multi-omics dataset of 122 HCCs, augment the (phospho)proteome data using neural networks and develop models for cellular classification from H&E-stained slides to extract detailed histopathological features. WP2: Inference of gene regulatory mechanisms in HCC: Perform single-cell RNA+ATAC-seq on 30 HCCs and 10 normal liver samples to construct cell type-specific gene regulatory networks (GRNs). WP3. Identification of tumor effectors from CRISPR screens: Optimise statistical frameworks to identify HCC effectors from CRISPR screen data. WP4. Prognostic model development: Develop machine learning (ML) models integrating multi-omics and pathology data to identify prognostic biomarkers, incorporating learnings from GRNs and tumor effectors. WP5: Validation of biomarkers: Validate biomarkers from the prognostic models in an external cohort. 1. Enriched multi-omics datasets with inferred (phospho)proteome profiles and digital pathology. 2. Comprehensive single-cell RNA+ATAC-seq data, revealing cell type-specific GRNs in HCC. 3. Optimized statistical framework for CRISPR screens, identifying HCC-specific effectors. 4. Robust ML models providing prognostic biomarkers validated in independent cohorts. This project aims to advance AI-driven precision oncology by integrating multi-omics data and computational pathology. The development of robust prognostic models will improve diagnostic accuracy and treatment personalization in HCC, potentially serving as a blueprint for other cancer types. These advancements promise to enhance patient care by enabling more informed clinical decisions and improving overall outcomes in oncology.