RiboDetector

RiboDetector identifies and removes ribosomal RNA (rRNA) sequences from metagenomic, metatranscriptomic, noncoding RNA (ncRNA), and ribosome profiling sequencing data to reduce rRNA-derived interference in downstream analyses.

Key Features:

• Bi-directional Long Short-Term Memory (BiLSTM) Neural Network: Leverages a BiLSTM neural network architecture to classify sequencing reads as rRNA or non-rRNA.
• High Accuracy and Low Misclassification Rate: Demonstrates at least a sixfold reduction in misclassification on benchmark datasets and produces false positives that are not biased toward specific Gene Ontology (GO) terms.
• Generalizability: Detects novel rRNA sequences with low sequence identity (<90%) to training sequences.
• Optimized Performance: Provides implementations optimized for GPU and CPU processing, able to process 40 million reads in under six minutes on a personal computer, with GPU mode ≈50× and CPU mode ≈15× faster than competing approaches.

Scientific Applications:

• Transcriptomics and translatomics: Reduces rRNA-derived reads in transcriptomic and translatomic datasets to enable more accurate downstream analyses.
• Metagenomic and metatranscriptomic studies: Removes rRNA from complex community sequencing datasets to decrease rRNA interference in environmental and microbiome analyses.
• ncRNA and ribosome profiling analyses: Facilitates analysis of noncoding RNA (ncRNA) and ribosome profiling datasets by minimizing rRNA contamination.
• Investigation of RNA activity and regulation: Supports analysis of RNA activity profiles and RNA-based regulatory mechanisms by reducing abundant rRNA signal.

Methodology:

Uses a Bi-directional Long Short-Term Memory (BiLSTM) neural network for rRNA read identification and provides optimized implementations for GPU and CPU execution.