AUTHORS
Beena Elizabeth Thomas, PhD, Karanpreet Bhatia, PhD2, Jon Rodiger3, Rita Mormando4, Alexander Gout, PhD5, Jonathan Keats, PhD6, Lawrence H Boise, PhD7, Benjamin Gabriel Barwick, PhD8, Vikas Gupta, MD, PhD9
AFFILIATIONS
1 Winship Cancer Institute of Emory University, Atlanta, Georgia, United States
2 Winship Cancer Institute of Emory University, Atlanta, Georgia, United States
3 Multiple Myeloma Research Foundation, Norwalk, Connecticut, United States
4 Multiple Myeloma Research Foundation, Norwalk, Connecticut, United States
5 Multiple Myeloma Research Foundation, Norwalk, Connecticut, United States
6 Tgen, Phoenix, Arizona, United States
7 Winship Cancer Institute of Emory University, Atlanta, Georgia, United States
8 Winship Cancer Institute of Emory University, Atlanta, Georgia, United States
9 Winship Cancer Institute of Emory University, Atlanta, Georgia, United States
ABSTRACT
Background. The Multiple Myeloma Research Foundation (MMRF) CoMMpass study has generated a rich, publicly available longitudinal multiple myeloma (MM) dataset that integrates clinical, cytogenetic, demographic, and genomic data including RNA-sequencing, whole genome sequencing, and Exome sequencing, as well as paired baseline-to-progression samples. However, most analyses leverage only a subset of this dataset, focusing on a single molecular classification system, cohort-level survival models, or single disease timepoint. Much of MM's biological complexity lies in how gene expression changes based on molecular subtype, cytogenetics, patient characteristics, and disease stage, and how it changes at relapse. These multidimensional associations remain underexplored because interrogating them requires substantial bioinformatics expertise and custom analyses.
Platform. MyelomaScape is a point-and-click R Shiny application for interactive exploration of the multidimensional CoMMpass IA24 dataset (1,043 patients clinically annotated; 806 with RNA-sequencing data; 756 in the baseline survival analysis cohort; 84 with paired baseline-progression samples). MyelomaScape enables interrogation across multiple orthogonal analytical dimensions: (i) three molecular classification frameworks in parallel—cytogenetic lesions, UAMS 7-class (Zhan 2006), and Skerget 12-class (2024)—evaluating the same hypothesis across molecular contexts; (ii) demographic and clinical axes; race, age, ISS, LDH, beta-2-microglobulin, and treatment-incorporated as stratification factors and covariates, enabling race-aware and stage-aware modeling that cohort-average analyses omit; (iii) longitudinal analysis, tracking within-patient expression change from diagnosis to progression in the paired cohort; and (iv) interaction analyses testing whether a gene's prognostic effect is dependent on for e.g., subtype or lesion, via formal interaction models and joint two-gene permutation testing. Additional modules provide adjusted Cox and continuous/spline modeling, gene-ratio and gene-signature scoring, covariate screen (screen gene sets for association with outcome), a signature-independence test (prognostic beyond, e.g., proliferation?), internal validation (ROC, cross-validation, bootstrap), and external validation of any interaction in an independent cohort (GSE2658, n=559). All plots and csv files are downloadable.
Illustrative findings. As a representative application, we used the platform to survey the BCL-2 apoptotic gene family for cytogenetic-subtype-dependent prognostic effects. Among these, BBC3-which encodes the canonical pro-apoptotic BH3-only protein PUMA-stood out: despite its pro-death function, BBC3 was associated with improved survival exclusively in high-risk t(4;14)+ MM (pooled HR 0.33, 95% CI 0.20–0.55, p<0.001), with no effect elsewhere (HR 1.10). This subtype-specific association was confirmed by formal interaction testing (p=0.001; permutation p=0.004, adjusted for age, race, ISS, LDH, beta-2-microglobulin, treatment and six cytogenetic lesions), replicated in both CoMMpass (HR 0.39) and GSE2658 (HR 0.27), and remained significant after adjustment for del(17p) (HR 0.42, p=0.009), indicating independence from p53 deletion. Because EP300 acetylates p53 to induce PUMA, we then tested a BBC3/EP300 ratio as a readout of this axis; it was prognostic in CoMMpass (HR 0.54, p=0.019) and GSE2658 (HR 0.64, p=0.043); the same ratio did not replicate when the subgroup was defined by UAMS MS class instead, illustrating how conclusions can depend on the classification framework used. Separately, the platform identified a STAT5B × t(11;14) interaction (NEG/POS HR ratio 2.32, 95% CI 1.41–3.82, p<0.001). Higher STAT5B was associated with worse outcome in t(11;14)-negative patients (HR 1.54 per IQR, 95% CI 1.24–1.92, p<0.001) but not in t(11;14)-positive patients (HR 0.58, 95% CI 0.34–1.00). This interaction was not externally validated as significance was not reached in GSE2658 for either group. This explicit evidence-grading is itself a platform feature.
Conclusions. By exposing the molecular, demographic, longitudinal, and interaction dimensions of CoMMpass data through an interactive interface with built-in external validation and evidence grading, MyelomaScape lets investigators move beyond cohort-level associations to the subgroup-specific relationships masked in conventional analyses in minutes rather than weeks without programming expertise.
DISCLOSURES
Beena Elizabeth Thomas, PhD: No relevant conflicts of interest to declare.