4mCNLP-Deep

4mCNLP-Deep predicts N4-methylcytosine (4mC) modification sites in DNA sequences using deep learning with convolutional neural networks and word embedding–based sequence representations.

Key Features:

• Convolutional Neural Network Model: Uses convolutional neural networks (CNNs) to classify DNA sequence regions as 4mC or non-4mC sites.
• Word Embedding Representation: Encodes DNA sequences using a word embedding approach based on vector representations generated from k-mer tokens.
• 3-mer word2vec Encoding: Utilizes a 3-mer corpus with word2vec modeling to convert nucleotide sequences into numerical feature vectors for model training.
• Performance Evaluation on C. elegans Dataset: Demonstrates predictive performance on a Caenorhabditis elegans genome dataset with reported metrics including ACC 0.9354, MCC 0.8608, Sp 0.8996, Sn 0.9563, and AUC 0.9731.
• K-mer Feature Exploration: Incorporates k-mer sequence analysis to evaluate feature representations for 4mC site prediction.

Scientific Applications:

• Epigenetic Modification Analysis: Identifies N4-methylcytosine modification sites involved in epigenetic regulation of gene expression.
• Genome Annotation: Supports annotation of DNA methylation patterns within genomic sequences.
• Functional Genomics Studies: Investigates relationships between 4mC modifications and biological processes such as chromosome stability and cellular development.

Methodology:

4mCNLP-Deep encodes DNA sequences using 3-mer word2vec embeddings, trains convolutional neural networks on labeled 4mC and non-4mC sequences, and evaluates predictive performance using k-fold cross-validation.