Tranquillyzer: A Flexible Neural Network Framework for Structural Annotation and Demultiplexing of Long-Read...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Long-read single-cell RNA sequencing (scRNA-seq) offers full-length transcriptome analysis.
• High error rates and complex library structures in long-read scRNA-seq challenge cell barcode (CBC) and unique molecular identifier (UMI) identification.
• Existing methods struggle with data variations like shifts, truncations, and non-canonical ordering.

Purpose of the Study:

• To develop a robust and flexible deep learning framework for processing long-read scRNA-seq data.
• To accurately identify CBCs and UMIs despite sequencing errors and library variability.
• To provide a scalable solution for analyzing large-scale long-read transcriptomic datasets.

Main Methods:

• Introduced Tranquillyzer, a deep learning framework utilizing a hybrid neural network architecture.
• Employed a global, context-aware design for precise identification of structural elements.
• Enabled rapid, one-time model training for custom library formats.

Main Results:

• Tranquillyzer accurately identifies CBCs and UMIs, even with shifted, degraded, or repeated elements.
• The framework demonstrates flexibility in supporting established and custom single-cell protocols.
• Model training is efficient, typically completed within hours on standard GPUs.

Conclusions:

• Tranquillyzer provides a flexible, scalable, and accurate solution for long-read scRNA-seq data processing.
• The framework overcomes limitations of existing methods in handling data complexities.
• It facilitates reliable demultiplexing and deduplication for enhanced transcriptomic analysis.