Minimally invasive testing is essential for early cancer detection, impacting patient survival rates significantly. Our study aimed to establish a pioneering cell-free immune-related miRNAs (cf-IRmiRNAs) signature for early cancer detection. We analyzed circulating miRNA profiles from 15,832 participants, including individuals with 13 types of cancer and control. The data was randomly divided into training, validation, and test sets (7:2:1), with an additional external test set of 684 participants. In the discovery phase, we identified 100 differentially expressed cf-IRmiRNAs between the malignant and non-malignant, retaining 39 using the least absolute shrinkage and selection operator (LASSO) method. Five machine learning algorithms were adopted to construct cf-IRmiRNAs signature, and the diagnostic classifies based on XGBoost algorithm showed the excellent performance for cancer detection in the validation set (AUC: 0.984, CI: 0.980-0.989), determined through 5-fold cross-validation and grid search. Further evaluation in the test and external test sets confirmed the reliability and efficacy of the classifier (AUC: 0.980 to 1.000). The classifier successfully detected early-stage cancers, particularly lung, prostate, and gastric cancers. It also distinguished between benign and malignant tumors. This study represents the largest and most comprehensive pan-cancer analysis on cf-IRmiRNAs, offering a promising non-invasive diagnostic biomarker for early cancer detection and potential impact on clinical practice.
Keywords: Cell-free immune-related miRNAs; Early detection of cancers; Machine learning algorithms; Pan-cancer analysis.
© 2024. The Author(s).