A machine learning model for cancer screening in dogs using comprehensive circulating microRNA profiles.

Abstract

BACKGROUND: MicroRNAs (miRNAs) are non-coding RNAs involved in cancer-related biological processes. To date, no studies have determined that liquid biopsy using miRNA can specifically identify dogs with cancer from a mixed population of dogs with and without non-malignant diseases. HYPOTHESIS/OBJECTIVES: To assess the utility of a diagnostic model that differentiates dogs with cancer from a combined group of healthy dogs and dogs with non-malignant diseases, using miRNA profiles obtained by next-generation sequencing (NGS) and analyzed using machine learning. ANIMALS: A total of 574 dogs were enrolled in the study: 168 with cancer, 138 with non-malignant diseases, and 268 healthy controls. METHODS: Plasma samples from all dogs were analyzed by NGS to generate comprehensive miRNA profiles. Models were developed using DataRobot, based on the 50 most highly expressed miRNAs. The optimal model was selected based on area under the curve (AUC) results obtained using 5-fold cross-validation. RESULTS: The miRNA-based model accurately distinguished dogs with cancer from those without cancer, achieving an AUC of 0.907, with both sensitivity and specificity of 0.85. CONCLUSIONS AND CLINICAL IMPORTANCE: A model integrating NGS-derived miRNA profiles with machine learning can serve as a diagnostic approach for cancer detection in dogs. Such a model can distinguish dogs with cancer from both healthy dogs and those with non-malignant disease. These findings suggest that such a model could be used as a screening test for dogs with cancer in veterinary practice.