Regulatory association prediction of microRNA and cancer radiation response based on Bayesian personalized ranking matrix factorization and deep auto-encoder

Radiotherapy, a primary treatment for cancer, seeks to deliver a lethal dose to the tumor while sparing surrounding healthy tissue. Cumulative evidence indicates that microRNAs (miRNAs) participate in the regulation of tumor radiation response. Therefore, characterizing the regulatory association between miRNA and cancer radiation response is essential for increasing the efficacy of radiotherapy. Traditional biological experiments often face the challenge of low efficiency and high failure rate. Here, we proposed the method, called BPRMCRA, to predict potential miRNA-cancer radiation response regulatory associations based on biological experimental data obtained from published literature. Firstly, we employed the Bayesian personalized ranking matrix factorization method to decompose the potential features from the regulatory association matrix composed of biological experimental data. Secondly, we introduced relevant features of miRNAs and cancer from a biological perspective and utilized the deep auto-encoder to decrease the feature dimensions of miRNAs and cancer. Finally, the soft voting method was applied for the prediction of regulatory association. The results demonstrate that the BPRMCRA method robustly outperforms other methods, and is therefore a very valuable approach for accurate prediction of the miRNA-cancer radiation response regulatory associations.