A classification model for lncRNA and mRNA based on k-mers and a convolutional neural network

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Long-chain non-coding RNA (lncRNA) plays a crucial role in biological activities.
• Distinguishing lncRNA from messenger RNA (mRNA) is challenging due to sequence similarities.
• Effective identification models for lncRNA and mRNA are essential.

Purpose of the Study:

• To develop a robust classification model for differentiating lncRNA and mRNA sequences.
• To leverage k-mer frequency distribution and convolutional neural networks for improved RNA classification.
• To determine the optimal k-mer combinations for accurate lncRNA/mRNA identification.

Main Methods:

• Transformation of lncRNA and mRNA sequences into k-mer frequency matrices.
• Screening of species-specific k-mers using relative entropy.
• Development of a convolutional neural network (CNN) model for sequence classification.
• Comparative analysis against traditional machine learning algorithms.

Main Results:

• The proposed CNN model achieved the highest classification accuracy in humans, mice, and chickens.
• Optimal classification was observed using 1-mers, 2-mers, and 3-mers.
• The model's recognition ability was validated on single sequences.

Conclusions:

• A novel classification model for lncRNA and mRNA based on k-mers and CNNs was successfully established.
• The developed model significantly outperforms Random Forest, Logistic Regression, Decision Tree, and Support Vector Machine.
• High classification accuracies (e.g., 0.9872 in humans, 0.9963 in chickens) highlight the model's effectiveness.