Linking gene expression data with patient survival times using partial least squares

Area of Science:

• Bioinformatics
• Biostatistics
• Genomics

Background:

• Integrating genotypic and phenotypic data is crucial for patient stratification.
• Analyzing patient survival data with censoring presents significant statistical challenges.
• Existing methods often do not fully leverage complex genotypic data for survival analysis.

Purpose of the Study:

• To develop a robust statistical method for analyzing patient survival data with censored outcomes.
• To effectively link high-dimensional genotypic data with phenotypic information, specifically survival times.
• To address the limitations of traditional regression models when dealing with censored survival data.

Main Methods:

• Reformulation of survival analysis with censored data into a Poisson regression problem.
• Integration of Partial Least Squares (PLS) for handling high-dimensional, collinear genotypic data.
• Application of Generalized Linear Regression (GLR) to accommodate various response variable types, including survival times.

Main Results:

• The proposed method successfully circumvents issues associated with censored data.
• Identified linear combinations of variables correlate strongly with patient survival times while accounting for covariate variability.
• The algorithm demonstrates computational efficiency, avoiding matrix decompositions during iterations.

Conclusions:

• The combined PLS and GLR approach offers an effective solution for analyzing censored survival data in the presence of large genotypic datasets.
• This method provides a valuable tool for linking genetic information with patient outcomes, applicable to various cancer studies.
• The algorithm's speed and effectiveness were validated on lung carcinoma and diffuse large B-cell lymphoma datasets.