DemuxTrans: Transformer and temporal convolution network for accurate barcode demultiplexing in nanopore sequencing

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Oxford Nanopore Technologies (ONT) direct RNA sequencing (dRNA-seq) provides high-resolution, single-molecule analysis.
• Current multiplex barcoding methods for dRNA-seq lack robustness and struggle with accurate demultiplexing of nanopore signals.
• Existing methods fail to capture both local patterns and long-range dependencies in sequencing data, limiting accuracy and efficiency.

Purpose of the Study:

• To develop an advanced solution for precise barcode demultiplexing in nanopore sequencing.
• To enhance the accuracy, efficiency, and adaptability of multiplexed RNA sample identification.
• To overcome limitations in current demultiplexing approaches for dRNA-seq.

Main Methods:

• A hybrid deep learning framework named DemuxTrans was developed.
• DemuxTrans integrates Multi-Layer Feature Fusion, Transformers, and Temporal Convolutional Networks (TCN).
• The framework is designed to capture both local sequence patterns and long-range dependencies.

Main Results:

• DemuxTrans achieved state-of-the-art performance across multiple datasets.
• The model demonstrated high accuracy, recall, and F1-score in barcode demultiplexing.
• DemuxTrans effectively balances local feature extraction, global context modeling, and long-term dependency capture.

Conclusions:

• DemuxTrans offers a scalable and efficient solution for barcode demultiplexing in nanopore sequencing.
• The framework enables precise identification of multiplexed RNA samples.
• This advancement improves throughput for transcriptomic and epigenomic analyses.