Cargando…

Predict Colon Cancer by Pairing Plasma miRNAs: Establishment of a Normalizer-Free, Cross-Platform Model

BACKGROUND: Plasma miRNAs are emerging biomarkers for colon cancer (CC) diagnosis. However, the lack of robust internal references largely limits their clinical application. Here we propose a ratio-based, normalizer-free algorithm to quantitate plasma miRNA for CC diagnosis. METHODS: A miRNA-pair ma...

Descripción completa

Detalles Bibliográficos
Autores principales: Qin, Da, Guo, Qingdong, Wei, Rui, Liu, Si, Zhu, Shengtao, Zhang, Shutian, Min, Li
Formato: Online Artículo Texto
Lenguaje:English
Publicado: Frontiers Media S.A. 2021
Materias:
Acceso en línea:https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8101326/
https://www.ncbi.nlm.nih.gov/pubmed/33968711
http://dx.doi.org/10.3389/fonc.2021.561763
Descripción
Sumario:BACKGROUND: Plasma miRNAs are emerging biomarkers for colon cancer (CC) diagnosis. However, the lack of robust internal references largely limits their clinical application. Here we propose a ratio-based, normalizer-free algorithm to quantitate plasma miRNA for CC diagnosis. METHODS: A miRNA-pair matrix was established by pairing differentially expressed miRNAs in the training group from GSE106817. LASSO regression was performed to select variables. To maximize the performance, four algorithms (LASSO regression, random forest, logistic regression, and SVM) were tested for each biomarker combination. Data from GSE106817 and GSE112264 were used for internal and external verification. RT-qPCR data acquired from another cohort were also used for external validation. RESULTS: After validation through four algorithms, we obtained a 4-miRNA pair model (miR-1246 miR-451a; miR-1246 miR-4514; miR-654-5p miR-575; miR-4299 miR-575) that showed good performance in differentiating CC from normal controls with a maximum AUC of 1.00 in internal verification and 0.93 in external verification. Tissue validation showed a maximum AUC of 0.81. Further external validation using RT-qPCR data exhibited good classifier ability with an AUC of 0.88. CONCLUSION: We established a cross-platform prediction model robust against sample-specific disturbance, which is not only well-performed in predicting CC but also promising in the diagnosis of other diseases.