Machine learning-driven programmed cell death signature for prognosis and drug candidate discovery in diffuse large B-cell lymphoma: Multi-cohort study and experimental validation.

BACKGROUND

Relapse and drug resistance are major contributor to chemotherapy failure in diffuse large B-cell lymphoma (DLBCL). Programmed cell death (PCD), a key mechanism in tumor progression and resistance, has emerged as a promising biomarker for predicting prognosis and chemotherapy sensitivity in DLBCL.

MATERIALS AND METHODS

This study integrated 15 PCD patterns and RNA-seq data from 3428 DLBCL patients (eight cohorts). PCD Score (PCDS) was developed using 101 machine learning algorithm combinations. Using PCDS, patients were stratified into high/low-risk groups through integrated bioinformatics analyses. The antitumor activity of candidate agents was validated through CCK-8, dual Hoechst 33342/Annexin V-PI apoptosis assays, and xenograft models, demonstrating tumor-suppressive efficacy.

RESULTS

A 17-gene PCDS developed by machine learning demonstrated high prognostic accuracy across cohorts, with high-risk patients showing significantly worse survival (P < 0.001). PCDS was integrated with clinical features to construct a nomogram with high predictive performance. Enrichment analysis showed upregulated proliferation pathways and suppressed immune/cell adhesion pathways in high-risk group, with increased Tregs and decreased cytotoxic CD8+ T cells (activated/effector memory subsets) and NK cells (P < 0.05). High-risk patients showed reduced sensitivity to standard chemotherapy (cyclophosphamide/doxorubicin/vincristine). Network pharmacology predicted Phloretin and Parthenolide as high-risk-specific therapeutic agents, with in vitro validation confirming their antitumor activity (Phloretin: 80.77 μM; Parthenolide: 0.93 μM). Furthermore, Parthenolide exhibited high sensitivity against DLBCL cells. Subsequent in vitro and in vivo experiments demonstrated its efficacy in inducing apoptosis and suppressing tumor growth in xenograft models. Enrichment analysis showed downregulation of the Phagosome, Lysosome, and Antigen processing and presentation pathways in the high-risk group, which were upregulated following treatment with Phloretin and Parthenolide. These findings that they may inhibit tumor progression by regulating these pathways.

CONCLUSION

The PCDS effectively predicts the post-chemotherapy prognosis of DLBCL patients. Moreover, Phloretin and Parthenolide exhibit promising potential as therapeutic agents for high-risk DLBCL patients with poor prognosis.