Optimized splitting of mixed-species RNA sequencing data

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Gene expression studies using mixed-species models (e.g., human and mouse xenografts/co-cultures) are vital for understanding cellular dynamics in development and disease.
• Challenges in analyzing mixed-species RNA sequencing data arise from similar mRNA sequences between species, impacting accurate transcript quantification.

Purpose of the Study:

• To identify optimal strategies for analyzing mixed-species RNA sequencing data.
• To evaluate the performance of alignment-dependent and alignment-independent methods for species-specific read classification.

Main Methods:

• Evaluation of alignment-dependent methods, including alignment to a pooled reference index followed by re-alignment to individual genomes.
• Assessment of alignment-independent methods, such as convolutional neural networks, for classifying RNA sequencing reads based on conserved sequence patterns.

Main Results:

• Alignment to a pooled reference index, with optimal alignment for species classification and subsequent re-alignment to individual genomes, achieved high accuracy across various species ratios.
• Alignment-independent methods, like convolutional neural networks, demonstrated over 85% accuracy in classifying RNA sequencing reads.
• Both approaches performed well with different human and mouse read ratios, but mixed-genome alignment with optimized read separation showed lower error rates.

Conclusions:

• Optimized alignment-based strategies provide a more accurate approach for transcript quantification in mixed-species RNA sequencing data compared to non-alignment methods.
• Effective species-specific read partitioning is achievable with both alignment-dependent and alignment-independent methods, but traditional alignment offers superior precision.