Stepwise classification of cancer samples using clinical and molecular data

Area of Science:

• Bioinformatics
• Statistical modeling
• Computational biology

Background:

• Integrating clinical and molecular data can enhance classifier accuracy.
• Current integrative tools lack efficiency and can overshadow subtle data contributions.
• Existing methods are often impractical and costly due to universal data requirements.

Purpose of the Study:

• To develop a novel, cost-efficient stepwise classification method for integrating clinical and molecular data.
• To address limitations of existing integrative classifiers, such as coarse data combination and high costs.
• To improve prediction accuracy while minimizing resource expenditure.

Main Methods:

• A stepwise classification approach was developed, leveraging distinct predictive powers of clinical and molecular data.
• Classification algorithms were applied independently to clinical data first, then molecular data if prediction uncertainty exceeded a threshold.
• The method adaptively determines the proportion of samples requiring molecular data based on expected accuracy gains.

Main Results:

• The stepwise classifier demonstrated adaptive behavior, adjusting molecular data usage based on predictive needs.
• Experimental results confirmed the approach's efficiency in utilizing high-dimensional molecular data.
• The method achieved comparable or superior accuracy to classifiers using only clinical or molecular data.

Conclusions:

• The novel method provides a cost-efficient classifier with performance on par with or exceeding single-data-type approaches.
• It avoids unnecessary molecular tests for many individuals, potentially reducing diagnosis waiting times and patient distress.
• The stepwise classification method is implemented in the R-package stepwiseCM, available on the Bioconductor website.