Hyb4mC

Hyb4mC predicts DNA N^4-methylcytosine (4mC) sites to enable analysis of epigenetic regulation processes such as DNA replication initiation, mismatch repair, and transposon inactivation.

Key Features:

• DNA2vec sequence embedding: Applies DNA2vec for sequence embedding to capture comprehensive sequence information for downstream modeling.
• Hyb_Caps (Capsule Neural Network): Employs a capsule neural network subnet to generalize from limited 4mC samples and model spatial hierarchies in sequence features.
• Hyb_Conv (Attention-enhanced Convolutional Neural Network): Integrates an attention mechanism with a text convolutional neural network to enhance feature learning and focus on relevant sequence patterns.
• Multi-species prediction: Provides prediction capability across multiple species to address disparate sample abundances among species.

Scientific Applications:

• Epigenetics research: Predicts 4mC sites to support studies of DNA methylation-mediated regulation.
• Gene regulation analysis: Enables investigation of mechanisms such as DNA replication initiation, mismatch repair, and transposon inactivation via predicted 4mC sites.
• Comparative multi-species studies: Supports cross-species comparison of 4mC distribution where sample numbers vary between species.

Methodology:

Performs DNA2vec sequence embedding followed by a dual-subnet architecture consisting of Hyb_Caps (capsule neural network) and Hyb_Conv (attention-enhanced text convolutional neural network), with benchmark evaluations against recently proposed methods.