Novel liquid biopsy based on comprehensive small RNA profiling for screening of cancer in dogs utilizing machine learning

Recent investigations have demonstrated the relevance in cancer of a variety of small RNAs, beyond microRNAs (miRNAs). However, the comprehensive application of these RNAs to cancer detection in dogs remains unexplored.

We investigated whether combining miRNAs and other small RNAs using a custom database improves the accuracy of machine learning cancer screening in dogs compared with models using only miRNAs.

A total of 389 dogs were enrolled: 85 with cancer, 77 with nonmalignant diseases, and 227 healthy controls.

Small RNAs in blood samples were sequenced, and miRNAs (via miRBase) and other small RNAs (via a custom database) were profiled. Based on these profiles, 2 datasets were prepared: one with only miRNAs and another combining miRNAs with other small RNAs. For each dataset, 12 classification models were constructed using identical methods to enable paired comparisons of performance, with 5-fold cross-validation and area under the receiver operating characteristic curve (AUC) as the primary metrics.

The model that integrated both miRNAs and other small RNAs yielded a significantly higher AUC compared with the miRNA-only model (median AUC: 0.926 [95% CI, 0.911-0.944] vs 0.881 [0.860-0.897]; Wilcoxon signed-rank test, P < .001), indicating enhanced capability in discriminating malignant tumors in dogs.

We investigated the comprehensive integration of diverse small RNA species, beyond miRNAs, using a machine learning approach for cancer screening in dogs. Although further prospective validation in independent cohorts is needed, our results indicate that this strategy improves the performance of cancer classification models in dogs.