Machine learning-based single-sample molecular classifier for cancer grading.

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Morphological tumor grading is crucial for cancer treatment and prognosis.
• Intermediate tumor grades often lack prognostic significance due to interobserver variability.
• A need exists for more reliable methods to assess tumor grade and patient risk.

Purpose of the Study:

• To develop a molecular-based classifier for predicting high- and low-grade tumors.
• To enable accurate risk assessment using gene expression data from single samples.
• To improve prognostic capabilities for breast, lung, and renal cancers.

Main Methods:

• Devised a molecular classifier using gene expression data (RNA sequencing or microarray).
• Developed a preprocessing procedure requiring only single-sample expression data.
• Validated the classifier's performance on both RNA-seq and microarray data.

Main Results:

• Molecular grades (mGrades) strongly correlated with histological grades and clinical stage.
• mGrades effectively assessed risk levels for intermediate (G2) tumor samples.
• Identified common and unique biological/genetic features associated with low and high mGrades across cancer types.

Conclusions:

• The molecular classifier provides a reliable method for tumor risk stratification.
• mGrades enhance prognostic accuracy, especially for intermediate-grade tumors.
• Gene expression patterns offer valuable insights for cancer research and diagnostics.