Thyroid Cancer Serum Proteomics and Machine Learning Model

A 2026 study by Da Zhang presents a diagnostic model for thyroid cancer using serum proteomics and ensemble machine learning. The model was developed using serum samples from 414 thyroid cancer patients and 430 healthy controls analyzed via MALDI-TOF MS. It identifies 12 key peptide peaks and demonstrates high diagnostic accuracy.

Use Cases

• Train ensemble machine learning models on serum proteomic data from 414 patients and 430 controls for thyroid cancer classification.
• Identify key diagnostic peptide peaks from MALDI-TOF MS data using SHAP or LIME for feature importance analysis.
• Validate a simplified diagnostic model based on the 12 identified peptide peaks against single-protein biomarkers.
• Perform GO and KEGG enrichment analysis on differentially expressed proteins to explore immune regulation and lipid metabolism pathways in thyroid cancer.

Strengths

• Dataset includes a substantial cohort of 844 total serum samples (414 patients, 430 controls).
• Model interpretability is supported by SHAP and LIME analyses, identifying 12 key peptide peaks.
• Diagnostic performance is rigorously evaluated using ROC-AUC and Decision Curve Analysis (DCA).

Limitations

• The dataset is very small in file size (12.3 KB), suggesting it contains summary results or a model description, not the raw proteomic data.
• Data originates from a single-center cohort, which may limit generalizability to other populations.
• The primary file format is a DOCX document, which is not a standard format for structured machine learning data.

Provenance

Source

figshare, authored by Da Zhang.

Collection Method

Serum samples analyzed using MALDI-TOF mass spectrometry; machine learning models applied for diagnostic construction.

Freshness

Last updated March 25, 2026.