Machine learning-driven investigation on liquid-liquid phase separation-related prognostic signature in diffuse large...

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Diffuse large B-cell lymphoma (DLBCL) is the most common aggressive non-Hodgkin lymphoma, exhibiting significant heterogeneity.
• Current risk stratification methods for DLBCL may not fully capture the disease's complexity.

Purpose of the Study:

• To develop and validate a prognostic model based on liquid-liquid phase separation-related genes (LRGs) for improved DLBCL risk stratification.
• To explore the biological and microenvironmental differences between risk groups identified by the model.

Main Methods:

• Analysis of transcriptomic and clinical data from four independent cohorts (n=768).
• Application of multiple machine learning algorithms to identify prognostic LRGs and construct a 6-LRG predictive model.
• Validation using survival analysis, time-dependent ROC curves, and multivariable modeling.

Main Results:

• The 6-LRG model successfully stratified DLBCL patients into distinct overall survival groups across all datasets.
• The model demonstrated robust predictive performance with high Area Under Curve (AUC) values at 1, 3, and 5 years.
• The 6-LRG model proved independent of established clinical variables and enhanced risk prediction when integrated into a nomogram.

Conclusions:

• The 6-LRG model offers valuable prognostic information for DLBCL, potentially refining risk stratification.
• The model highlights distinct biological and immune characteristics associated with different risk groups, suggesting underlying mechanisms.
• Prospective validation in larger cohorts is recommended for clinical implementation.