MTTLm6A: A multi-task transfer learning approach for base resolution mRNA m6A site prediction based on an improved transformer

We propose MTTLm6A, an improved transformer-based multi-task transfer learning approach for predicting base resolution m6A sites in Saccharomyces cerevisiae. First, the RNA sequences are encoded using one-hot encoding. Then, we construct a multitask model that combines a convolutional neural network (CNN) with a Multi-Head Attention deep framework. This model not only detects low-resolution m6A sites but also assigns reasonable probabilities to the predicted sites. Finally, we employ transfer learning to predict base-resolution m6A sites based on the low-resolution m6A sites. Experimental results on Saccharomyces cerevisiae m6A data demonstrate that MTTLm6A achieves an area under the receiver operating characteristic (AUROC) of 77.13% and outperforms state-of-the-art models. To enhance user convenience, we have made a user-friendly web server for MTTLm6A publicly available at http://47.242.23.141/MTTLm6A/index.php.

Fig. 1 The diagram of the model. The source domain stage model is used to discriminate low-resolution m6A sites from non-m6A sites and the target domain stage model is used to identify high-resolution m6A sites from low-resolution m6A sites.

Fig. 2 The internal structure of the model.

The Dataset Used in Our Paper

m6A:

The training set, validation set for building model:
train_unbalance_group(zip)

The independent test set for assessment predictor:
test_unbalance_group(zip)

See the text of the paper for more detail information.