KNOWLEDGE-INFORMED ML FOR CANCER DAIGNOSIS AND PROGNOSIS

Cancer remains one of the leading causes of mortality worldwide, creating an urgent need for improved diagnostic and prognostic techniques that support early detection and personalized treatment strategies. Conventional cancer diagnosis relies on imaging, histopathology, and limited molecular biomarkers, which often fail to capture the complex biological mechanisms driving tumor development and progression. Recent advances in highthroughput sequencing technologies have enabled comprehensive multi-omics profiling, including mRNA expression, microRNA regulation, and DNA methylation data, providing deeper insights into tumor biology. However, traditional machine learning approaches typically treat these datasets as independent features, ignoring biological relationships such as gene interactions, signaling pathways, and protein networks. This limitation reduces predictive performance and interpretability in clinical applications. To address these challenges, this work proposes a KnowledgeInformed Machine Learning (KIML) framework that integrates biological knowledge with multiomics patient data for cancer diagnosis and prognosis prediction. The proposed system constructs a biological knowledge graph using curated biomedical databases such as Gene Ontology, KEGG pathways, and protein–protein interaction networks. Graph-based learning models are then employed to capture relationships between genes, pathways, and molecular signals while incorporating patient-specific omics information. By embedding structured biological knowledge into the learning process, the system improves predictive accuracy and enhances model interpretability. Experimental evaluation demonstrates that the proposed approach outperforms conventional machine learning techniques in classification accuracy and prognostic risk stratification. The integration of domain knowledge with machine learning provides a promising direction for developing intelligent clinical decision-support systems that assist oncologists in identifying high-risk patients and designing targeted treatment strategies.