AI-derived five-gene signature predicts risk in multiple myeloma under bortezomib-based therapy

Multiple myeloma (MM) is a plasma cell malignancy characterized by clonal expansion within the bone marrow (BM) and remains largely incurable despite recent therapeutic advances. While prognostic indicators have predominantly focused on tumor-intrinsic features, the BM tumor microenvironment (TME) plays a critical role in disease progression and therapeutic resistance. Here, we present a robust prognostic model derived from transcriptomic profiling of the CD138-negative BM fraction, enabling precise risk stratification of patients undergoing bortezomib-based induction therapy. Using bioinformatic deconvolution and a custom immune-stromal signature matrix, we identified gene expression patterns representative of the MM TME. Through rigorous feature selection and elastic-net penalized Cox regression, we developed the MM-5C model, consisting of five genes (SOX11, METTL11B, C3, RBM10, and HOMEZ), which stratifies patients into biologically distinct risk groups and demonstrates prognostic independence from established cytogenetic and clinical staging systems. This model underscores the pivotal role of TME components in shaping therapeutic outcomes and offers a scalable, clinically translatable tool for personalized risk stratification. Our findings highlight the necessity of integrating microenvironmental insights into MM prognostication and pave the way for microenvironment-informed therapeutic decision-making.