Identifying New Potential Biomarkers in Adrenocortical Tumors Based on mRNA Expression Data Using Machine Learning

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Adrenocortical carcinoma (ACC) is a rare endocrine malignancy with poor prognosis.
• Previous multi-omics studies identified molecular subtypes (C1A, C1B) linked to patient survival.

Purpose of the Study:

• To apply machine learning (ML) to transcriptome data for unbiased classification of ACC.
• To identify novel prognostic marker genes in ACC.

Main Methods:

• Utilized publicly available TCGA-ACC transcriptome data (n=79).
• Applied Uniform Manifold Approximation and Projection (UMAP) for clustering.
• Employed random-forest-based learning for marker gene identification and validation on a secondary dataset.

Main Results:

• UMAP clustering yielded two distinct groups (ACC-UMAP1, ACC-UMAP2) correlating with known C1B and C1A clusters.
• Identified potential new prognostic marker genes, including SOAT1 and EIF2A1.
• Validated findings on an independent dataset, noting a correlation between benign tumors and the good prognosis ACC-UMAP1/C1B cluster.

Conclusions:

• ML approaches effectively re-identified and refined known prognostic ACC subgroups.
• Discovered novel potential prognostic marker genes for adrenocortical carcinoma.
• Findings contribute to a better understanding of ACC molecular heterogeneity and prognosis.