Enhanced early detection of thyroid cancer using ensemble machine learning and serum proteomics

Area of Science:

• Biochemistry
• Oncology
• Bioinformatics

Background:

• Thyroid cancer diagnosis is challenging due to asymptomatic onset and limited specificity of current imaging and biomarker methods.
• Early detection is crucial for improving patient prognosis, especially after metastasis.

Purpose of the Study:

• To develop and validate a diagnostic model integrating serum proteomics and machine learning for early thyroid cancer detection.
• To identify key serum peptides indicative of thyroid cancer and assess the model's clinical utility.

Main Methods:

• Serum samples from 414 thyroid cancer patients and 430 controls were analyzed using MALDI-TOF MS.
• Multiple machine learning algorithms were employed to construct and validate diagnostic models.
• SHAP and LIME analyses were used for model interpretability, and key peptides were identified via feature importance.

Main Results:

• The integrated machine learning model achieved excellent discriminative performance.
• Twelve peptide peaks significantly associated with thyroid cancer were identified, forming the basis for a simplified, accurate diagnostic model.
• Enrichment analysis indicated these peptides are involved in immune regulation and lipid metabolism.

Conclusions:

• A validated serum peptide-based diagnostic model integrating machine learning shows promise for non-invasive early thyroid cancer detection.
• This approach offers a potential improvement over single-biomarker strategies and warrants further research.