Integrated multiomics analysis and machine learning refine molecular subtypes and prognosis for thyroid cancer

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Thyroid cancer (THCA) presents significant molecular heterogeneity, complicating accurate prognosis and personalized treatment.
• Current prognostic models often utilize limited data and algorithms, impacting their reliability and clinical utility.

Purpose of the Study:

• To classify molecular subtypes of THCA using multiomics data.
• To develop a robust prognostic model for THCA by integrating multiple machine learning algorithms and omics layers.
• To evaluate the model's performance and its correlation with immunogenomic features and drug sensitivity.

Main Methods:

• Integration of five omics layers from THCA patients.
• Application of eleven clustering algorithms to identify molecular subtypes.
• Development of a prognostic model (Consensus Machine Learning-Driven Signature - CMLS) using stable prognosis-related genes (SPRGs) and 99 machine learning combinations.
• Validation across internal and external patient cohorts.

Main Results:

• Identification of three distinct molecular subtypes (CS1-CS3), with CS2 demonstrating the poorest prognosis.
• Establishment of a nine-gene CMLS model with strong prognostic capability.
• Correlation of low-CMLS group with better outcomes, enhanced immune infiltration, higher tumor mutational burden (TMB) and tumor neoantigen (TNB) load, and improved immunotherapy response.
• Identification of six potential therapeutic agents for high-CMLS patients.

Conclusions:

• The study presents a robust, multiomics-driven classification of THCA.
• The developed CMLS model offers clinical relevance for prognostic prediction and personalized therapy guidance in THCA.
• Findings support improved risk stratification and tailored treatment strategies for thyroid cancer patients.