Revealing heterogeneity in chronic lymphocytic leukemia: AI-driven insights into aggressive and indolent disease subtypes.

Abstract

e19029 Background: Chronic lymphocytic leukemia (CLL) exhibits a broad spectrum of clinical behaviors, from indolent courses requiring minimal intervention to aggressive forms demanding immediate treatment. The underlying heterogeneity of CLL poses significant challenges in predicting disease progression and tailoring patient-specific therapeutic strategies. Leveraging advanced and novel artificial intelligence (AI) and machine learning (ML) methodologies, our study aims to dissect the complexity of CLL, focusing on the identification and characterization of biomarkers that differentiate aggressive and indolent disease forms. Methods: The NetraAI is a priorietaryAI/ML system that is also being utilized in a research collaboration at a leading American oncology institute (not disclosed) that integrates machine intelligence, augmented intelligence, and an agent that interacts with AI-derived patient representations. This system was designed to learn from datasets with a small number of samples and many independent variables. This system operates by generating hypotheses in the form of interactive representations of patient populations that reveal heterogeneity along with statistically significant driving factors. Finally, an automated agent called AutoPlay interacts with these representations and generates an accounting of all variables, subpopulations of samples, and it also generates a human readable perspective on what was discovered by the data set. Results: We applied the NetraAI to a CLL dataset and used it to refine the analysis and interpretation of the corresponding transcriptomic data (GSE39411). Significantly lower levels of FADS3, GSDME, LPL, IMMT, NMB, and AEBP1 and higher expressions of COBLL1, P2RY1R, PDE8A, SYNE2, and FCRL3 were identified as hallmarks of indolent CLL, suggesting these markers could inform less aggressive disease management strategies. Furthermore, within a total of 104 CLL samples, we uncovered two distinct subpopulations within aggressive CLL, delineated by unique genetic markers. Group A, encompassing 31 aggressive CLL samples is predominantly characterized by the expression of lipoprotein lipase (LPL), while Group B, comprising 22/23 aggressive samples, is characterized by ZBTB20 and SYNE2. Our poster will also provide glimpses into a more refined heterogeneity that emerged from the analysis. Our results not only underscore the heterogeneity within aggressive CLL, but also highlights the potential for these markers to guide prognostic assessments and therapeutic decisions. Conclusions: This study not only advances our understanding of CLL heterogeneity, but also sets the stage for the development of more personalized and effective treatment modalities. By identifying drivers of disease aggression and indolence, we pave the way for targeted therapies that could significantly improve patient outcomes in CLL.