Sparse integrative dictionary learning resolves conserved drug-response states in multiple myeloma (MM)

Area of Science:

• Hematology
• Computational Biology
• Genomics

Background:

• Multiple myeloma (MM) is a plasma-cell malignancy with significant mortality despite advances.
• Relapse is common, suggesting shared adaptation mechanisms across diverse therapies.
• Existing studies often analyze drug responses individually, missing cross-therapy conserved programs.

Purpose of the Study:

• To develop an interpretable framework for identifying conserved transcriptional adaptation programs in multiple myeloma across different treatments.
• To analyze transcriptional shifts induced by various therapies to uncover shared response patterns.
• To generate mechanistic hypotheses regarding treatment adaptation and drug resistance in MM.

Main Methods:

• Developed PRISM-MM, a sparse integrative dictionary-learning framework for analyzing drug-induced transcriptional changes.
• Incorporated single-cell RNA sequencing (scRNA-seq) for cell-state-level interpretability.
• Applied PRISM-MM to a multi-study MM perturbation compendium and validated findings in external datasets.

Main Results:

• Identified conserved transcriptional response programs across diverse MM treatments.
• Validated these programs in held-out bulk and external scRNA-seq datasets.
• Observed recurrent remodeling characterized by reduced secretory activity and increased inflammatory/adhesion states.
• Highlighted a RFXAP-centered immune-regulatory axis potentially driving adaptation.

Conclusions:

• PRISM-MM provides an interpretable method for discovering conserved drug-response programs in MM.
• Conserved programs reveal common adaptation strategies, including immune-interface state enrichment.
• Identified potential therapeutic targets and mechanistic insights into MM drug resistance.