Transcriptomic data in tumor-adjacent normal tissues harbor prognostic information on multiple cancer types.

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Traditionally, tumor-adjacent normal tissues are used for comparison, not as primary subjects for cancer prognostic marker identification.
• Differential gene expression analysis between tumor and normal tissues is a common precursor to prognostic analysis.
• Recent findings question the prognostic significance of differentially expressed genes (DEGs) in certain cancers.

Purpose of the Study:

• To investigate the prognostic value of transcriptomic data from both tumor and adjacent normal tissues.
• To compare the efficacy of normal tissue, tumor tissue, and DEG data in predicting cancer prognosis.
• To evaluate the utility of machine learning and feature selection in identifying prognostic markers from normal tissues.

Main Methods:

• Utilized The Cancer Genome Atlas (TCGA) dataset for transcriptomic analysis.
• Employed Cox regression models for prognostic analysis.
• Applied machine learning models and feature selection methods for survival prediction.

Main Results:

• Adjacent normal tissues demonstrated a higher proportion of prognostic genes for kidney, liver, and head and neck cancers.
• Normal tissues yielded superior survival prediction performance compared to tumor tissues and DEGs in machine learning models.
• External dataset validation confirmed that genes selected from adjacent normal tissues had higher prediction accuracy.

Conclusions:

• Gene expression levels in adjacent normal tissues represent significant and potentially overlooked prognostic markers.
• Rethinking the role of adjacent normal tissues could improve cancer prognosis prediction.
• The study highlights the prognostic potential of non-tumorigenic tissue in cancer research.