Classification based on extensions of LS-PLS using logistic regression: application to clinical and multiple genomic...

Area of Science:

• Bioinformatics
• Genomic Data Analysis
• Biostatistics

Background:

• High-dimensional genomic data analysis often excludes valuable clinical information.
• Integrating clinical and genomic data is suggested to enhance predictive accuracy.
• Existing prediction methods frequently analyze genomic or clinical data in isolation.

Purpose of the Study:

• To develop and evaluate classification methods that simultaneously utilize both clinical and genomic data.
• To apply dimensionality reduction techniques specifically to high-dimensional genomic variables.
• To compare the predictive performance of integrated approaches against single-data-type methods.

Main Methods:

• Proposed one-step classification approaches based on extensions of the least squares-partial least squares (LS-PLS) method for logistic regression.
• Utilized partial least squares (PLS) for dimensionality reduction of genomic data.
• Compared prediction performances through simulations and analysis of real cancer study datasets.

Main Results:

• Methods relying solely on clinical or genomic data generally showed poor performance.
• The proposed LS-PLS methods demonstrated advantages for classification tasks.
• Prediction results were encouraging and stable across different datasets and feature selections.

Conclusions:

• Simultaneous analysis of clinical and genomic data, particularly with LS-PLS extensions, offers improved predictive power.
• The developed methods provide stable and reliable predictions for classification tasks.
• The R package lsplsGlm facilitates the application of these advanced classification techniques.