2026 ASH®
Annual Meeting & Exposition December 12-15, 2026 ▪ New Orleans, Louisiana

Paper Details

Abstracts, authors, affiliations, and disclosures.

Sunday, December 13, 2026
06:00 p.m. - 08:00 p.m.
MCCNO - Hall F
Access, Representation, Population Health, Technology and Procedures, Assays, Imaging, Serologic Tests, Emerging Technologies, Profiling, Treatment Considerations, Drug Development, Biological Therapies, Antibody Therapy, Bispecific Antibody Therapy, Checkpoint Inhibitor, Immunotherapy, Adverse Events, Artificial Intelligence (AI), Machine Learning, Biological Processes, CHIP, Genomics, Immune Mechanism, Microbiome, Multi-Systemic Interactions, Diseases, Lymphoid Malignancies, Lymphomas, Plasma Cell Disorders, Myeloid Malignancies, Chronic Myeloid Malignancies, Measurable Residual Disease, Non-Hodgkin Lymphoma, Research, Translational Research, MPN, Clinical Research, Clinical Trials, Health Outcomes Research, Health Disparities Research, Real-World Evidence, Study Population, Human, Adult
PAPER

3956 Machine learning-derived blood-based immune signatures stratify progression risk in precursors of multiple myeloma

Nayda Bidikian, MD, Noe Perron, PhD2, Rosa Toenges, MD3, Mark Bal, MD4, Tomoki Iemura, MD, PhD5, Romanos Sklavenitis-Pistofidis, MD PhD6, Francesco Corrado, MD7, Anushka Ghosh, PhD8, Dor David Abelman, PhD(c), HBHSc9, Floris Chabrun, MD10, Lorena Pantano, PhD11, Jacqueline Perry, MPH12, Hannah Ashton13, Hector Gracia14, Esperanza Martin-Sanchez, PhD, MSc, BSc15, Julia Colchie, BA, MA16, Jean Regard, PhD17, Ana Batlle Lopez, MD, PhD18, Tarek Mouhieddine, MD19, Elizabeth O'Donnell, MD20, Omar Nadeem, MD21, Bruno Paiva, PhD22, Gaddy Getz, PhD23, Irene M Ghobrial, MD24
1 Broad Institute of MIT & Harvard, Cambridge, Massachusetts, United States, Dana-Farber Cancer Institute, Boston, Massachusetts, United States, Harvard Medical School, Boston, United States, Yale New Haven Hospital, New Haven, Connecticut, United States
2 Broad Institute of MIT & Harvard, Cambridge, Massachusetts, United States, Dana-Farber Cancer Institute, Boston, Massachusetts, United States, Harvard Medical School, Boston, United States
3 Broad Institute of MIT & Harvard, Cambridge, Massachusetts, United States, Dana-Farber Cancer Institute, Boston, Massachusetts, United States, Harvard Medical School, Boston, United States
4 Broad Institute of MIT & Harvard, Cambridge, Massachusetts, United States, Dana-Farber Cancer Institute, Boston, Massachusetts, United States, Harvard Medical School, Boston, United States
5 Broad Institute of MIT & Harvard, Cambridge, Massachusetts, United States, Dana-Farber Cancer Institute, Boston, Massachusetts, United States, Harvard Medical School, Boston, United States
6 Broad Institute of MIT & Harvard, Cambridge, Massachusetts, United States, Dana-Farber Cancer Institute, Boston, Massachusetts, United States, Harvard Medical School, Boston, United States
7 Broad Institute of MIT & Harvard, Cambridge, Massachusetts, United States, Dana-Farber Cancer Institute, Boston, Massachusetts, United States, Harvard Medical School, Boston, United States
8 Broad Institute of MIT & Harvard, Cambridge, Massachusetts, United States, Dana-Farber Cancer Institute, Boston, Massachusetts, United States, Harvard Medical School, Boston, United States
9 Broad Institute of MIT & Harvard, Cambridge, Massachusetts, United States, Dana-Farber Cancer Institute, Boston, Massachusetts, United States, Harvard Medical School, Boston, United States
10 Dana-Farber Cancer Institute, Boston, Massachusetts, United States, University Hospital of Angers, Angers, France, University of Angers, Angers, France
11 Broad Institute of MIT & Harvard, Cambridge, Massachusetts, United States, Dana-Farber Cancer Institute, Boston, Massachusetts, United States, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, United States
12 Broad Institute of MIT & Harvard, Cambridge, Massachusetts, United States, Dana-Farber Cancer Institute, Boston, Massachusetts, United States, Harvard Medical School, Boston, United States
13 Broad Institute of MIT & Harvard, Cambridge, Massachusetts, United States, Dana-Farber Cancer Institute, Boston, Massachusetts, United States, Harvard Medical School, Boston, United States
14 Clínica Universidad de Navarra, Madrid, Spain
15 Clínica Universidad de Navarra, Madrid, Spain
16 Broad Institute of MIT & Harvard, Cambridge, Massachusetts, United States, Dana-Farber Cancer Institute, Boston, Massachusetts, United States, Harvard Medical School, Boston, United States
17 Broad Institute of MIT & Harvard, Cambridge, Massachusetts, United States, Dana-Farber Cancer Institute, Boston, Massachusetts, United States, Harvard Medical School, Boston, United States
18 Broad Institute of MIT & Harvard, Cambridge, Massachusetts, United States, Dana-Farber Cancer Institute, Boston, Massachusetts, United States, Harvard Medical School, Boston, United States
19 Dana-Farber Cancer Institute, Boston, Massachusetts, United States
20 Dana-Farber Cancer Institute, Boston, Massachusetts, United States
21 Dana-Farber Cancer Institute, Boston, Massachusetts, United States
22 Clínica Universidad de Navarra, Madrid, Spain
23 Broad Institute of MIT & Harvard, Cambridge, Massachusetts, United States, Harvard Medical School, Boston, United States, Krantz Family Center for Cancer Research, Boston, Massachusetts, United States
24 Broad Institute of MIT & Harvard, Cambridge, Massachusetts, United States, Dana-Farber Cancer Institute, Boston, Massachusetts, United States, Harvard Medical School, Boston, United States

INTRODUCTION

Risk stratification in smoldering multiple myeloma (SMM) currently relies on bone marrow (BM) assessment of tumor burden and cytogenetics, an invasive procedure limiting serial monitoring. Immune dysfunction is implicated in progression from monoclonal gammopathy of undetermined significance (MGUS) and SMM to multiple myeloma (MM), but its temporal evolution and predictive value remain unclear. To address this gap, we applied machine learning (ML) to single-cell immune profiling of peripheral blood (PB), a minimally invasive source, to identify stage-specific immune programs associated with progression across the MGUS - MM spectrum. 

METHODS

We performed 5' single-cell RNA and T-cell receptor (TCR) sequencing of baseline PB samples from 48 patients with MGUS (n=20 progressors to SMM) and 64 patients with SMM (n=24 progressors to MM), followed longitudinally for a median follow-up of 3.5 years. The dataset comprised 337,000 cells across 32 immune subsets, pseudobulked per patient within stage and adjusted for age, sex, M-spike, and pool batch. Differentially expressed genes (DEGs) between progressors and non-progressors were identified per cell type by covariate-adjusted pseudobulk analysis with Benjamini-Hochberg (BH) correction. We trained a ridge-logistic regression machine learning (ML) model to perform cell-type-resolved panel search over the locked DEG pool across all T-cell subsets. Model performance was assessed using joint permutation family-wise error correction and bootstrap-derived 95% confidence intervals (CI).

RESULTS

To define immune changes preceding clinical progression, we compared stage-matched progressors and non-progressors at two transitions, MGUS-to-SMM and SMM-to-MM. At MGUS-to-SMM, future progressors showed a selective immunosuppressive program specifically in CD16+ non-classical monocytes (NCM). DEG analysis identified an upregulation of the tumor-permissive prostaglandin PGE2 pathway (reflected in PTGS2/HPGD balance, BH p = 0.012), the inhibitory Fc receptor FCGR2B (p = 0.033), and weakening of the USP18/ISG15 interferon feedback brake (p = 0.026) in progressors. By contrast, CD14+ classical monocytes from the same patients showed increased scores in interferon (IFN)-response and inflammasome-related gene signatures, indicating that MGUS-to-SMM progression is not driven by global myeloid dysfunction but by a selective immunosuppressive shift restricted to the NCM subset.

At SMM-to-MM, the most coordinated changes occurred within the T-cell compartment and were distinct from the canonical exhaustion phenotype described in overt MM. Across all T-cell subsets, progressors showed reduced IFN-response, oxidative phosphorylation, and cytotoxic-effector gene signatures, together with rising cell-cycle transcription that increased near progression. Notably, canonical T-cell exhaustion signatures decreased rather than increased across all T-cell subsets (Wherry β = -0.04, q = 0.017; Sade-Feldman β = -0.03, q = 0.017), and TCR repertoire diversity was preserved, with rarefied diversity metrics trending toward broader repertoires (Shannon β = +0.33, Richness β = +27.6, both q = 0.034). Together, these findings suggest that SMM-to-MM progression is associated with coordinated impairment of T-cell effector and homeostatic programs, without evidence of increasing T-cell exhaustion signatures.

Finally, to translate these findings into a clinical biomarker, we trained a ML model to identify cell-type-resolved gene signatures across all T-cell subsets. After joint-permutation family-wise correction, a compact multi-gene GZMK+ CD8+ T effector memory (TEM) signature emerged within the top-performing panels and stratified patients with SMM at risk of progression with an AUC of 0.85 (95% CI 0.75 to 0.92). Additional T-cell subset signatures achieved AUCs > 0.80, indicating that progression-associated T-cell dysfunction is represented across the compartment and not confined to a single subset. Interestingly, these signatures did not predict progression in the MGUS cohort, highlighting their stage specificity. 

CONCLUSION

Our findings define a temporally structured immune trajectory preceding MM and identify stage-specific immune alterations that can be translated into a blood-based ML-derived biomarker for progression. These results support minimally-invasive PB single-cell immune profiling as a strategy for risk stratification and longitudinal monitoring of MM precursor disease.
Nayda Bidikian, MD: No relevant conflicts of interest to declare.