Integrative analysis of cancer prognosis data with multiple subtypes using regularized gradient descent

Area of Science:

• Genomics
• Cancer Research
• Biostatistics

Background:

• High-throughput profiling studies aim to identify genes/single nucleotide polymorphisms (SNPs) linked to cancer prognosis.
• Different cancer subtypes may share common susceptibility genes, necessitating methods that account for this heterogeneity.

Purpose of the Study:

• To develop a novel analytical approach for identifying prognosis-associated genes/SNPs across multiple cancer subtypes.
• To model the genetic basis of cancer subtypes using a heterogeneity model with overlapping susceptibility genes.
• To apply an accelerated failure time (AFT) model for prognosis analysis.

Main Methods:

• A regularized gradient descent approach for gene-level analysis to identify genes with prognosis-associated SNPs.
• Utilizing a heterogeneity model to accommodate overlapping yet distinct gene/SNP sets for different subtypes.
• Employing an accelerated failure time (AFT) model to analyze prognosis data.

Main Results:

• The proposed regularized gradient descent approach demonstrates superior performance compared to alternatives, yielding more true positives and fewer false positives in simulations.
• Analysis of non-Hodgkin lymphoma (NHL) prognosis data identified distinct sets of genes associated with DLBCL, FL, and CLL/SLL subtypes.
• The developed method achieved the best prediction performance on the NHL dataset.

Conclusions:

• The novel approach effectively identifies prognosis-associated genes/SNPs across cancer subtypes, accounting for shared genetic factors.
• This method offers improved accuracy and computational efficiency for cancer prognosis studies.
• The findings highlight the utility of the heterogeneity model and gradient descent for dissecting complex genetic architectures in cancer.