Prediction of N6-methyladenosine sites using convolution neural network model based on distributed feature...

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• N-methyladenosine (m6A) is a prevalent mRNA modification crucial for cellular processes like development, RNA splicing, and cancer.
• Existing computational models for m6A site prediction have yielded unsatisfactory results.
• There is a critical need for efficient and accurate computational tools to identify m6A sites.

Purpose of the Study:

• To develop an intelligent and highly discriminative computational model for predicting m6A sites.
• To leverage natural language processing techniques for automatic feature extraction of m6A motifs.
• To improve the accuracy and efficiency of m6A site identification compared to existing methods.

Main Methods:

• Utilized word2vec from natural language processing to represent m6A motifs as numerical descriptors.
• Extracted features automatically from the human genome without predefined motifs.
• Employed a convolution neural network (CNN) model for prediction using the extracted feature space.

Main Results:

• The m6A-word2vec model achieved high accuracy: 83.17% on dataset S1, 92.69% on S2, and 90.50% on S3.
• Performance was evaluated using a 10-fold cross-validation test.
• The developed model demonstrated superior performance compared to existing computational approaches.

Conclusions:

• The m6A-word2vec model is an effective computational tool for accurate m6A site discrimination.
• This model offers a practical solution for advancing elementary and pharmaceutical research, including drug design.
• The approach provides a novel method for automatic motif discovery and prediction in epigenetics.