Abstract 2056: Development of a response prediction Model for venetoclax-based therapy utilizing molecular interactions among BCL2 family members

Abstract Venetoclax (VEN), in combination with hypomethylating agents (HMAs) such as azacitidine or decitabine, is a widely used treatment for AML patients, yet the molecular biomarkers contributing to sensitivity to the VEN+HMA regimen remain incompletely elucidated, particularly with regard to the role of BCL2-family proteins. In this study, we retrospectively collected BMMCs or PBMCs from AML patients prior to initiation of VEN+HMA therapy at Winship Cancer Institute of Emory University. A total of 47 specimens were analyzed, including 33 from newly diagnosed AML patients and 14 from relapsed or refractory AML cases at the time of sample collection. Clinical outcomes were categorized based on the best response to VEN+HMA therapy, with refractory cases labeled as "NR" (non-responders; 17 patients) and all other cases labeled as "R" (responders; 26 patients). To investigate protein interactions between the BCL2 family, we employed the Single-molecule Protein Interaction Detection (SPID) platform (Chun et al., Nat. Biomed. Eng., 8, 2024). 12 distinct protein biomarkers were assessed based on three anti-apoptotic proteins: BCL2, BCLxl, and MCL1. ROC analysis was performed to evaluate the association between individual biomarkers and clinical outcomes of VEN+HMA therapy. To develop a predictive model incorporating BCL2-family biomarkers, we utilized a linear model rectified with a ReLU activation function. The trained model was interpreted using SHAP value analysis. To enhance the variability of biomarkers, we calculated normalized protein complex levels relative to protein levels. For example, the BCL2-BAX complex level was divided by the BCL2 protein level to assess the fraction of occupied BCL2 across samples. Among the 21 biomarkers analyzed, the normalized BCLxl-BAK complex level demonstrated the strongest predictive performance, with an AUC of 0.734 (95% CI: 0.536-0.859) for distinguishing responders to VEN+HMA therapy. We developed a response prediction model for VEN+HMA therapy with 12 biomarkers, and achieved performance with 96.2% sensitivity and 100% specificity. Interpretation of the model using SHAP analysis revealed key contributors to sensitivity and resistance to the VEN+HMA regimen. Specifically, Higher levels of the BCL2-BAX and BCLXL-BAK complexes were associated with sensitivity to VEN+HMA therapy. Conversely, lower levels of the BCL2-BIM and MCL1-BIM complexes were linked to resistance to the treatment. Notably, the expression level of MCL1 (measured after heat denaturation) significantly influenced sensitivity to VEN+HMA therapy. This study highlights the potential of BCL2-family interaction profiles in developing a response prediction model for VEN+HMA sensitivity. Further model refinement and validation in independent cohorts are essential to improve its predictive performance and reliability. Citation Format: Byungsan Choi, Yunseo Lee, Lilly Gu, Keonho Lee, Eunyoung Kim, Martha L. Arellano, William G. Blum, Hongwon Lee, Janghee Woo. Development of a response prediction Model for venetoclax-based therapy utilizing molecular interactions among BCL2 family members [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2056.