Simulating the restoration of normal gene expression from different thyroid cancer stages using deep learning

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Thyroid cancer (THCA) is the most common endocrine malignancy with increasing incidence.
• Understanding molecular differences across THCA stages is crucial for diagnosis, prognosis, and treatment.
• Transcriptome State Perturbation Generator (TSPG) aids in identifying gene expression changes between sample types.

Purpose of the Study:

• To utilize TSPG to analyze gene expression in THCA tumors at different stages.
• To identify characteristic gene expression patterns associated with specific THCA stages.
• To uncover potential molecular targets for THCA diagnosis and treatment.

Main Methods:

• Applied TSPG to bulk RNA expression data from THCA tumors across progressive stages.
• Perturbed tumor expression data towards normal thyroid tissue transcriptional patterns.
• Analyzed perturbations to identify consistently up- or down-regulated genes and functions.

Main Results:

• SLC6A15 was down-regulated in stages 1-3 THCA, consistent with its tumor suppressor role.
• PLA2G12B was up-regulated in all samples, potentially linking thyroid metabolism to cancer.
• REN was up-regulated in late-stage (3-4) THCA, suggesting a role in angiogenesis and progression.
• Olfactory receptor activity was enriched in up-regulated genes, aligning with known cancer links.

Conclusions:

• TSPG is effective for analyzing large gene expression datasets and identifying class-specific differences.
• Identified specific genes perturbed in THCA, particularly in late-stage tumors.
• Provided evidence for potential stage-specific transcriptional signatures in THCA, offering future research targets.