Large language models completely understand molecular characteristics of squamous cervical cancer

Abstract Squamous cervical cancer (SCC) is a major cause of death in women, yet its molecular characteristics are poorly understood. Here, we profiled histopathological and molecular alterations in SCC, and used large language models (LLMs) for interpretation, reasoning, and understanding of multi-modal data. We implemented an immersive-knowledge prompting (iKLP) strategy to trigger LLMs, which interpreted 17.8%-20.3% of omic alterations known to be associated with cancer. Also, the emergence of cross-disciplinary reasoning in LLMs helped for interpreting phenotypic effects of SCC molecular alterations, exemplified by a prognostic biomarker HRG, and targetable kinases, CDK18 and CDK9. With experimental validations, LLM-reasoning showed >2-fold increased confidence for 68.5% of analyzed molecules. Strikingly, LLMs understood the information flow of cell-cell communications, and uncovered a CDK18-mediated immune escaping axis in orchestrating the crosstalk of malignant, immune, and niche cells. We anticipate that LLMs can be used for completely distinguishing between knowns and unknowns in any scientific problems.