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

Area of Science:

• Veterinary oncology
• Molecular diagnostics
• Bioinformatics

Background:

• MicroRNAs (miRNAs) are crucial non-coding RNAs in cancer development.
• Current diagnostic methods lack specificity in differentiating canine cancer from other conditions.
• Liquid biopsy using miRNAs for canine cancer detection remains largely unexplored.

Purpose of the Study:

• To develop and validate a diagnostic model for canine cancer detection.
• To differentiate dogs with cancer from healthy dogs and those with non-malignant diseases.
• To assess the utility of next-generation sequencing (NGS) and machine learning in canine cancer diagnostics.

Main Methods:

• Plasma samples from 574 dogs (168 cancer, 138 non-malignant disease, 268 healthy) were analyzed using NGS.
• Comprehensive miRNA profiles were generated and analyzed using machine learning (DataRobot).
• A model was built using the 50 most highly expressed miRNAs, optimized via 5-fold cross-validation.

Main Results:

• The miRNA-based diagnostic model achieved an area under the curve (AUC) of 0.907.
• The model demonstrated high accuracy, with both sensitivity and specificity at 0.85.
• It successfully distinguished dogs with cancer from healthy controls and dogs with non-malignant conditions.

Conclusions:

• A diagnostic model integrating NGS-derived miRNA profiles and machine learning is effective for canine cancer detection.
• This approach can differentiate cancer-affected dogs from healthy and non-malignant disease groups.
• The findings support the potential use of this model as a non-invasive screening test in veterinary settings.