4acCPred

4acCPred predicts genomic regions carrying the N4-acetyldeoxycytosine (4acC) DNA modification to enable analysis of its role in gene expression regulation.

Key Features:

• Weakly supervised multi-instance learning: Employs a weakly supervised framework that treats broad sequencing-identified regions (hundreds of bases) as bags of instances to learn locus-level signals for 4acC.
• Multi-instance deep neural network: Uses a multi-instance-based deep neural network architecture to predict 4acC-carrying regions from genomic sequences.
• High predictive accuracy: Reports mean areas under the ROC curve of 0.9877 for cross-validation and 0.9899 for independent testing across four datasets.
• Motif mining via model interpretation: Extracts sequence motifs from the trained model, with identified motifs consistent with existing biological knowledge.
• Targeted modification and organism: Focuses on N4-acetyldeoxycytosine (4acC), a modification found to be abundant in Arabidopsis and associated with actively transcribed genes.

Scientific Applications:

• Mapping 4acC distribution: Identifies genomic regions carrying 4acC to map its distribution across genomes.
• Transcriptional regulation studies: Enables analysis of the association between 4acC presence and actively transcribed genes to investigate regulatory roles.
• Epigenetics and plant biology: Supports investigations in plant epigenetic regulation, particularly in Arabidopsis, and may inform studies of similar modifications in other organisms.

Methodology:

Training a multi-instance-based deep neural network under a weakly supervised learning framework by treating broad regions as 'bags' of instances, followed by motif mining through model interpretation.