Classification of colon cancer patients into consensus molecular subtypes using support vector machines

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Colon cancer exhibits molecular heterogeneity, necessitating accurate tumor classification for effective treatment.
• The Consensus Molecular Subtypes (CMS) framework provides a molecular subtyping approach for colon cancer.
• Current RNA-sequencing (RNA-Seq) based CMS classification methods show limitations in sensitivity and specificity.

Purpose of the Study:

• To develop and optimize an RNA-Seq-based gene classifier for molecular subtyping of colon cancer patients into CMS groups.
• To identify subtype-specific and survival-associated genes for improved classification accuracy.

Main Methods:

• Subtype-specific and survival-associated genes were identified using the Fuzzy C-Means algorithm and log-rank test.
• Patient classification into CMS groups was performed using support vector machines with backward elimination.
• Optimization of the RNA-Seq-based classifier involved selecting 25 genes to minimize classification error.

Main Results:

• An optimized 25-gene signature was identified for colon cancer classification with minimal error rates.
• Classification performance was evaluated using metrics including precision, sensitivity, specificity, false discovery rate, and balanced accuracy.
• CMS3-associated genes demonstrated the highest specificity but lowest sensitivity, attributed to a small patient cohort for this subtype.

Conclusions:

• A gene list for colon cancer classification with minimized error rates has been established.
• The developed classifier shows promise for improving the accuracy of molecular subtyping in colon cancer.
• Further validation is warranted, particularly for subtypes with limited patient representation in the current study.