Construct prognostic models of multiple myeloma with pathway information incorporated

Area of Science:

• Hematology
• Bioinformatics
• Computational Biology

Background:

• Multiple myeloma (MM) is a plasma cell malignancy requiring personalized treatment strategies.
• Current prognostic models often overlook pathway-level information crucial for disease progression.
• Gene-level models have limitations in capturing the complexity of MM pathogenesis.

Purpose of the Study:

• To develop and validate novel prognostic models for MM using pathway information.
• To compare the predictive performance of pathway-based models against gene-based and existing models.
• To assess the impact of missing data on model accuracy and explore imputation methods.

Main Methods:

• Implemented pathway score methods (ssGSEA, GSVA, z-scores) and group lasso with 14 pathway data sources.
• Utilized microarray data (GSE136324) for internal validation and two external datasets for further testing.
• Investigated the robustness of models against missing data using imputation techniques.

Main Results:

• The group lasso model incorporating VAX pathway information (Vax(grp)) demonstrated superior predictive performance over gene-based models.
• Immune-related pathways, including VAX, were identified as strong predictors of MM.
• The Vax(grp) model showed enhanced resistance to missing values (<5%) with an integrated imputation method.

Conclusions:

• Pathway-based models, particularly using group lasso with immune pathway data, are effective alternatives for MM prognosis.
• The developed R package (MMMs) facilitates the application and validation of these advanced prognostic tools.
• This approach enhances personalized medicine by improving therapeutic decision-making in MM.