ReorientExpress

ReorientExpress predicts the 5'-to-3' orientation of cDNA long-read sequencing reads from nanopore and PacBio sequencing platforms without requiring reference genomes or transcriptomes by using deep neural networks.

Key Features:

• Reference-free orientation prediction: Performs orientation prediction for cDNA long reads without relying on reference genomes or transcriptomes.
• Deep neural network models: Employs deep neural networks (deep learning) to learn sequence patterns predictive of read orientation.
• Compatibility with nanopore and PacBio: Applies to long-read sequencing data generated on nanopore and PacBio platforms.
• Training on closely related species: Can be trained using sequences from closely related species to improve predictive accuracy.
• Read clustering strategies: Can utilize read clustering strategies to enhance orientation prediction when references are unavailable.
• Intrinsic sequence feature–based prediction: Predicts read orientation based on intrinsic sequence features of the reads.
• Benchmarking and evaluation: Performance has been assessed through benchmarking and analysis across experimental settings.

Scientific Applications:

• Long-read transcriptomics: Orients cDNA long reads to support transcriptome analyses from nanopore and PacBio data.
• Non-model organisms: Enables transcriptomic studies in non-model organisms or samples lacking comprehensive genomic references by operating reference-free.
• Reference-free analyses: Provides oriented reads for downstream transcript-level analyses without requiring additional experimental interventions.
• Diverse biological samples: Facilitates studies of diverse samples and organisms where traditional reference-based orientation methods are inadequate.

Methodology:

Constructs and applies deep neural network models trained on sequence data or clustered reads (including training on closely related species) to predict read orientation from intrinsic sequence features, with validation by benchmarking and analysis.