CircRNA-disease association prediction with an improved unbalanced Bi-Random walk

Abstract

Studies show that circular RNAs (circRNAs), a type of non-coding RNAs, play various roles in biological processes such as the formation and progression of many different diseases. For this reason, identifying potential circRNAs associated with diseases is vital for early diagnosis. Determining these relationships experimentally requires a long process and is also expensive. For this reason, computational models are being developed to determine the relationships between circRNA and diseases. In this study, we recommend a technique called Improved Unbalanced Bi-Random Walk (UBRW) to identify potential circRNAs associated with diseases. The commonly used 5-fold cross-validation (CV) technique and leave one-out cross-validation (LOOCV) technique were applied to verify the predictive ability of our technique. The area under curve (AUC) values calculated in 5-fold CV and LOOCV are 0.8910 and 0.9669, respectively. Case studies on the commonly occurring gastric cancer and breast cancer were conducted to further validate the predictive performance of our method. When the results were examined, it was seen that the prediction ability of the UBRW method was quite successful.