Precision prediction of venetoclax-azacitidine treatment efficacy in acute myeloid leukemia via integrative drug screening and machine learning.

Venetoclax-azacitidine (VEN/AZA) has transformed acute myeloid leukemia (AML) therapy, yet reliable predictors of response remain lacking. We employ a multidisciplinary strategy combining ex vivo drug sensitivity testing, transcriptomic profiling, functional assays, and clinical data to identify determinants of VEN/AZA response. Core genes consistently associated with responsiveness are validated through CRISPR-Cas9 screening, with silencing of BCL2L1 and PINK1 preferentially enhancing drug sensitivity. Building on these insights, we develop and validate an eight-gene random forest model (RF8) that achieves high accuracy across four independent cohorts (n = 498). RF8 distills the downstream effects of genetic alterations to assist in predicting treatment response and outperforms existing genetic mutation-based signatures. Moreover, RF8 scores show a nearly monotonic relationship with clinical response probabilities and survival outcomes, enabling precise patient stratification. These findings demonstrate the feasibility of integrating transcriptomic and drug-response data to guide VEN/AZA therapy, representing an advance toward personalized therapeutic strategies.