Predicting continuous values of prognostic markers in breast cancer from microarray gene expression profiles

Area of Science:

• Genomics
• Oncology
• Bioinformatics

Background:

• Current breast cancer prognostic markers rely on protein levels or tumor phenotype.
• Microarrays have shown potential in classifying binary breast cancer subtypes, such as estrogen receptor (ER)-alpha status.

Purpose of the Study:

• To investigate if tumor gene expression profiles encode information for conventional prognostic markers.
• To predict ER protein values and redefine ER-positive/negative cutoffs using gene expression data.
• To predict other prognostic parameters like S-phase fraction (SPF), histological grade, and DNA ploidy.

Main Methods:

• Analysis of 48 primary breast tumors from lymph node-negative patients using 6728-element cDNA microarrays.
• Application of artificial neural networks trained with gene expression data for prediction.
• Prediction of ER protein values, SPF, histological grade, and DNA ploidy status.

Main Results:

• Artificial neural networks successfully predicted ER protein values on a continuous scale.
• A gene expression profile-directed threshold was determined to redefine ER-positive and ER-negative classes.
• A consistent reciprocal relationship was observed in gene expression levels crucial for both ER and SPF prediction.

Conclusions:

• Tumor gene expression profiles can predict conventional prognostic markers in breast cancer.
• Gene expression data offers a more biologically relevant approach to defining prognostic marker cutoffs.
• Further research using gene expression profiles can enhance understanding and improve breast cancer prognostic markers.