Tiara

Tiara classifies eukaryotic and organellar sequences in metagenomic datasets to enable identification and separation of eukaryotic nuclear and organellar genomic fragments for downstream ecological and evolutionary analyses.

Key Features:

• Deep-Learning Approach: Tiara uses a PyTorch-powered deep-learning framework to perform sequence classification.
• Two-Step Classification Process: A first stage classifies sequences into archaea, bacteria, prokarya, eukarya, organelle, and unknown, and a second stage refines organellar sequences into plastidial and mitochondrial fractions.
• Organellar Sequence Classification: Correctly identifies and separates organellar sequences, distinguishing plastidial versus mitochondrial origins.
• Performance and Efficiency: Shows comparable performance to EukRep for prokaryotic sequence classification, superior performance for eukaryotic sequence classification, and reduced computational time in benchmarks.
• Implementation: Implemented in Python 3.8 and tested on Unix-based systems; released under an MIT license.
• Benchmark Version: Version 1.0.1 was used for benchmarking in comparative evaluations.

Scientific Applications:

• Eukaryotic Diversity Studies: Identification of eukaryotic nuclear and organellar genomes to support analyses of eukaryotic diversity and evolutionary relationships.
• Metagenomics Research: Separation of eukaryotic, prokaryotic, and organellar sequences within complex metagenomic samples for downstream taxonomic and functional analyses.

Methodology:

Deep-learning models implemented in PyTorch with a two-stage classification: first into archaea, bacteria, prokarya, eukarya, organelle, and unknown, then refinement of organellar sequences into plastidial and mitochondrial categories; implemented in Python 3.8 and tested on Unix-based systems.