Bream: an open-source deep learning framework for simultaneous base calling and DNA methylation detection on novel...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Nanopore sequencing offers simultaneous genetic and epigenetic information detection.
• Developing accurate, open-source computational models for non-ONT platforms is challenging.

Purpose of the Study:

• To introduce Bream, an open-source deep learning framework for base calling and DNA modification detection.
• To evaluate Bream's performance on a novel nanopore sequencing platform.

Main Methods:

• Bream integrates convolutional neural networks (CNN) and recurrent neural networks (RNN) for base calling.
• A bidirectional LSTM with attention mechanism is used for methylation detection.
• The framework was trained and validated on datasets from *A. thaliana*, *O. sativa*, and *D. melanogaster* using Qitan Technology's QCell-384 platform.

Main Results:

• Bream achieved base-calling accuracies of 89.38%-91.83%, comparable to ONT's R9.4 platform.
• High-performance methylation detection was observed, with an AUC-ROC of 0.98 on *D. melanogaster* data.
• Whole-genome CpG methylation frequency estimates strongly correlated with bisulfite sequencing data (Pearson's r ≥ 0.96).

Conclusions:

• Bream is a powerful, transparent, and adaptable tool for simultaneous base calling and methylation detection.
• The framework supports emerging nanopore sequencing platforms, fostering open innovation.
• Bream facilitates advanced genomic analysis on non-ONT platforms.