m6Anet

m6Anet detects N6-methyladenosine (m6A) modifications from nanopore direct RNA sequencing raw current signals using a neural network within a Multiple Instance Learning framework.

Key Features:

• Multiple Instance Learning framework: Employs a Multiple Instance Learning (MIL) framework to train on site-level labels while accounting for unannotated read-level modification states.
• Neural network-based signal interpretation: Uses a neural network architecture trained to interpret raw current signals from nanopore direct RNA sequencing for m6A detection.
• Performance and generalization: Demonstrates improved accuracy over existing computational methods, achieves accuracy comparable to experimental techniques, and generalizes across different cell lines without retraining model parameters.
• Stoichiometry capture: Provides read-level stoichiometry estimates to quantify modification rates.
• Transcriptome-wide application: Enables transcriptome-wide identification and quantification of m6A from a single direct RNA sequencing run.

Scientific Applications:

• Transcriptome-wide m6A mapping: Mapping N6-methyladenosine (m6A) sites across the transcriptome using direct RNA sequencing data.
• Comparative cell-line analysis: Comparative studies of m6A distribution across different cell lines without model retraining.
• Quantitative modification studies: Quantifying read-level stoichiometry to study dynamic regulation of RNA modifications.
• Functional investigation: Investigating functional implications and potential regulatory roles of m6A in gene expression.

Methodology:

Applies a neural network within a Multiple Instance Learning (MIL) framework to analyze raw current signals from nanopore direct RNA sequencing, infer site-level m6A presence despite missing read-level labels, and estimate read-level stoichiometry.