MMLmiRLocNet

Introduction


Inspired by multi-view multi-label learning strategy, we proposed a computational method named MMLmiRLocNet for predicting the subcellular localizations of miRNAs. The MMLmiRLocNet predictor extracts diverse feature representations of biological sequences through an investigation into lexical, syntactic, and semantic aspects. Specifically, it integrates lexical attributes derived from k-mer physicochemical profiles, syntactic characteristics obtained via word2vec embeddings, and semantic representations generated by RNABERT feature embeddings. Finally, the multi-view consensus features and specific features extract network constructed to extract consensus features and specific features from various perspectives, respectively. The full connection networks are utilized as the output module to predict the miRNA subcellular localization. The results show that the MMLmiRLocNet outperforms previous methods in terms of F1 score, subACC, and Accuracy, and achieves highly comparable performance with the help of multi-view consensus features and specific features extract network.

Figure.1 The framework of MMLmiRLocNet. MMLmiRLocNet takes a miRNA sequence as input, encoding lexical, syntactic, and semantic features for biological sequences. Following the feature embedding blocks, it utilizes a multi-view consensus feature extraction network to learn distinct weights for nucleotides associated with each subcellular localization. Lastly, a fully connected layer is employed to execute the multi-label classification task.


References

Upon the usage the users are requested to use the following citation:

Tao Bai and Bin Liu*.
MMLmiRLocNet: miRNA Subcellular Localization Prediction based on Multi-view Multi-label Learning for Drug design (Submitted)