SpBLRSR: Schatten p-norm Constrained Bounded Low-rank Subspace Recovery for Predicting N7-methylguanosine (m7G)-disease Associations

Abstract

Background: As an essential positively charged RNA modification, N7-methylguanosine (m7G) has been reported to be associated with multiple diseases including cancers. While transcriptome-wide m7G sites have been identified by high-throughput sequencing approaches, the disease-associated m7G sites are still largely unknown. Therefore, computational methods are urgently needed to predict potential m7G-disease associations, which is crucial for understanding the biosynthetic pathways of tumorigenesis at the epi-transcriptome layer. Objective: We hope to develop an effective computational method that can accurately predict the associations between m7G sites and diseases, and then to prioritizing candidate m7G sites for novel diseases. Method: In this article, we proposed a Schatten p-norm constrained bounded low-rank subspace recovery (SpBLRSR) method for m7G-disease association prediction. An m7G-disease block matrix was built to alleviate the sparseness during the association pattern discovery process. By incorporating the low-rank representation (LRR) model and sparse subspace clustering (SSC) model, SpBLRSR was designed to capture both the global and local structures of the association pattern. Results: Compared with the benchmark methods, SpBLRSR achieved the best performance in predicting associations between m7G sites and disease, and in prioritizing m7G sites for novel diseases. Then the robustness of Schatten p-norm in our method was further validated via a noise contamination experiment. Finally, case study of breast cancer was performed to elucidate the biological meaning of our method. Conclusion: SpBLRSR exploits the disease pathogenesis at the epitranscriptome layer by predicting potential m7A sites for disease.