Abstract
Comprehensive understanding of premalignant lesions (PMLs) represents a pivotal opportunity for cancer early detection and interception. Recently, advances in multi-omics technologies and artificial intelligence (AI) methods have provided unprecedented insights into PML-induced tumorigenesis. In this paper, we firstly catalog clinically recognized PMLs across 15 cancer types, emphasizing their epidemiological profiles and malignant transformation potentials. Then, we summarize recent intriguing discoveries and remaining challenges from bulk, single-cell, and spatial omics studies, highlighting how these omics technologies reveal the dynamic molecular, cellular, and spatial evolution from precancerous states to invasive malignancies. We further discuss network-based computational strategies for multi-omics integration and tumorigenesis trajectory inference, with applications of recent deep learning-based AI approaches. Finally, we highlight translational implications for PMLs, including developing high-precision early-diagnosis biomarkers and targeted pharmacological preventive strategies. Collectively, this paper underscores how the convergence of high-resolution multi-omics with sophisticated AI is poised to redefine PML research, enabling pan-cancer exceedingly-early risk stratification and pharmacological prevention.