Establishment and validation of a diagnostic model for cholangiocarcinoma based on LightGBM machine-learning...

Area of Science:

• Oncology
• Bioinformatics
• Genomics

Background:

• Cholangiocarcinoma (CCA) poses a significant mortality risk, often due to delayed diagnosis.
• Identifying reliable biomarkers is essential for early CCA detection and effective treatment strategies.

Purpose of the Study:

• To develop an optimal diagnostic model for Cholangiocarcinoma (CCA) utilizing machine learning algorithms.
• To identify key differentially expressed genes (DEGs) associated with CCA for diagnostic and therapeutic insights.

Main Methods:

• Analysis of gene expression profiles from CCA tumor and adjacent non-tumor tissues.
• Application of 11 machine-learning algorithms, including WGCNA, F-test, characteristic importance, and Lasso regression, to identify key DEGs (APOF, DIO1, OTC).
• Construction and evaluation of diagnostic models, with LightGBM identified as the optimal performer via ROC curve analysis.

Main Results:

• The optimal LightGBM model achieved an AUC of 0.84, with accuracy, precision, and recall of 0.80, 0.83, and 0.90, respectively.
• Key genes APOF, DIO1, and OTC were identified as significantly downregulated in CCA tissues.
• Analysis revealed immune cell infiltration imbalances and identified CCL16 as a chemokine involved in CCA immunoregulation.

Conclusions:

• The developed LightGBM model demonstrates high potential for accurate CCA diagnosis.
• The identified genes (APOF, DIO1, OTC) and chemokine (CCL16) offer novel targets for CCA diagnosis and immunotherapy.
• This research provides a foundation for advancing CCA diagnostic tools and therapeutic approaches.