isONform

isONform reconstructs reference-free transcriptomes from Oxford Nanopore Technologies (ONT) cDNA long-read sequencing data to predict isoforms without relying on genome annotations.

Key Features:

• Reference-Free Analysis: Operates without annotated genomes or high-quality references, enabling isoform prediction directly from ONT cDNA reads.
• Gene Graphs from Fuzzy Seeds: Constructs gene graphs using fuzzy seeds derived from ONT reads to represent transcript variability.
• Iterative Bubble Popping: Refines gene graphs through an iterative "bubble popping" procedure to resolve alternative paths and enhance isoform detection.
• High Sensitivity with Precision Trade-off: Achieves higher sensitivity for isoform discovery compared to other reference-free methods such as RATTLE, with a modest reduction in precision.
• Consistency with Annotation-Based Methods: Produces predictions that show substantial agreement with annotation-based tools such as StringTie2 on biological datasets.
• Validation Across Dataset Types: Has been validated on simulated, synthetic, and biological datasets.
• Application as Orthogonal Method: Can be used both to construct isoforms in poorly annotated organisms and to orthogonally validate reference-based transcriptome reconstructions.

Scientific Applications:

• Transcriptome reconstruction in non-model organisms: Enables isoform discovery in species lacking comprehensive genomic annotations using ONT long reads.
• Alternative splicing analysis: Facilitates identification and characterization of alternative splicing events from long-read cDNA data.
• Novel isoform discovery: Detects novel transcript isoforms directly from sequencing data without prior annotations.
• Validation of reference-based predictions: Serves as an orthogonal approach to corroborate isoform calls from annotation-dependent tools.

Methodology:

Builds gene graphs from ONT cDNA reads using fuzzy seeds and refines those graphs via an iterative bubble-popping algorithm to produce isoform predictions; validated on simulated, synthetic, and biological datasets.