Transcriptome Assembly at Single-Cell Resolution with Beaver.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) reveals cellular transcriptome heterogeneity.
• Accurate full-length transcript reconstruction is challenging due to scRNA-seq data sparsity and dropout rates.
• Existing meta-assembly methods struggle to balance consensus and cell-specific transcript information.

Purpose of the Study:

• To develop a novel cell-specific transcript assembler for short-read scRNA-seq data.
• To improve the accuracy and precision of full-length transcript reconstruction at single-cell resolution.
• To address the limitations of current meta-assembly and single-sample assembly approaches.

Main Methods:

• Developed Beaver, a cell-specific transcript assembler utilizing a transcript fragment graph.
• Implemented an efficient dynamic programming algorithm for identifying candidate full-length transcripts.
• Integrated two random forest models with 51 features to estimate transcript expression likelihood in individual cells.

Main Results:

• Beaver demonstrated substantial performance improvements over existing meta-assemblers (Aletsch, TransMeta, PsiCLASS) and single-sample assemblers (Scallop2, StringTie2).
• At equivalent sensitivity levels, Beaver achieved significantly higher precision: 32.0%-64.6% over meta-assemblers and 10.1%-67.0% over single-sample assemblers.
• Experiments on real and simulated Smart-seq3 scRNA-seq data validated Beaver's superior accuracy.

Conclusions:

• Beaver offers a robust solution for accurate cell-specific transcript assembly from scRNA-seq data.
• The developed method effectively balances consensus assembly with the identification of cell-specific transcriptional signatures.
• Beaver represents a significant advancement in analyzing cellular transcriptome heterogeneity using scRNA-seq data.