HLMethy

HLMethy assigns labels to m6A candidates using a multiple instance learning framework to improve identification and localization of N6-methyladenosine (m6A) modifications from noisy m6A-seq/MeRIP-seq peak data.

Key Features:

• Multiple Instance Learning Framework: Employs a multiple instance learning framework to infer instance-level labels from peak-level m6A-seq/MeRIP-seq data.
• Label Assignment from Noisy Data: Assigns labels to m6A candidates derived from noisy m6A-seq/MeRIP-seq peaks to enable finer-resolution inference.
• Two Training Strategies: Implements two distinct training strategies to develop and refine classification capabilities.
• Benchmarking with Single-Base Resolution: Validated using m6A sites with single-base resolution as benchmarks.
• Comparable Instance-Level Performance: Demonstrates performance comparable to existing instance-level predictors on single-base resolution benchmarks.
• Resolution Improvement: Targets the 100–200 nucleotide resolution limitation of MeRIP-seq/m6A-seq to improve localization and quantification of modified residues.
• Applicability to Other Modifications: Framework can be applied to other RNA modifications identified via peak-calling approaches.

Scientific Applications:

• m6A identification and characterization: Identification and characterization of N6-methyladenosine (m6A) modifications within RNA transcripts from MeRIP-seq/m6A-seq data.
• Localization and quantification: Enhancing localization toward single-base resolution and estimating the number of modified residues within transcript regions identified by peak-calling.
• Data quality and cost reduction: Improving the effective data quality of m6A-seq/MeRIP-seq analyses and reducing reliance on single-base resolution assays.
• Extension to other epitranscriptomic studies: Adaptation for analysis of other RNA modifications detected through peak-calling methods.

Methodology:

HLMethy applies a machine learning-based multiple instance learning framework with two distinct training strategies to assign labels to m6A candidates derived from noisy m6A-seq/MeRIP-seq peak data and is evaluated using single-base resolution m6A sites as benchmarks.