LongTron

LongTron evaluates spliced-alignment accuracy in long-read RNA sequencing data from Oxford Nanopore and PacBio CCS by simulating platform-specific error modes, classifying alignment errors, and comparing intron-chain calls to splice sites and external assemblies to support transcriptome and isoform analyses.

Key Features:

• Error Simulation and Classification: Simulates error modes specific to Oxford Nanopore and PacBio CCS to characterize challenges in long-read spliced alignments.
• Random Forest Classifier: Trains a Random Forest on simulated data to categorize long-read alignments into two error types or a non-error category.
• Integration with Aligners and Datasets: Processes alignments from minimap2 and evaluates datasets including NA12878 (Oxford Nanopore), PacBio CCS, and a PacBio SKBR3 cancer cell line dataset.
• Comparison and Validation: Compares intron chains from long-read alignments against individual splice sites, short read assemblies, and outputs from the FLAIR pipeline to identify splice-site and intron-chain discrepancies.

Scientific Applications:

• Genome annotation refinement: Improves the precision of splice-site and intron-chain evidence used for updating genome annotations.
• Isoform-level expression quantification: Informs isoform-level expression estimates by distinguishing erroneous from valid spliced alignments.
• Novel transcript and isoform discovery: Aids identification and validation of novel transcripts and isoforms by highlighting alignment-supported and unsupported splice junctions.

Methodology:

Simulates Oxford Nanopore and PacBio CCS error modes, trains a Random Forest classifier on the simulated alignments, classifies new long-read alignments into three categories, uses minimap2 alignments as input, compares intron chains to individual splice sites, short read assemblies, and FLAIR outputs, and employs a modified GffCompare for analysis and validation against the cited datasets.