EvalDNA

EvalDNA evaluates the quality of de novo genome assemblies using supervised machine learning to produce a standardized quality score without requiring a reference genome.

Key Features:

• Supervised machine learning (random forest regression): Integrates multiple assembly quality metrics into a single predictive quality score using a random forest regression model.
• Quality metrics calculation: Computes an extensive set of metrics from assembled sequences to capture continuity, completeness, and accuracy.
• Reference-free assessment: Assesses assembly quality without needing an existing reference genome.
• Cross-species applicability: Developed for mammalian genomes and applicable to bacterial genome assemblies.
• Comparative analysis capability: Produces a standardized score enabling direct comparison among assemblies from different species.

Scientific Applications:

• Ranking genome assemblers: Used to rank assembler performance such as in evaluations of GAGE human chromosome 14 assemblies.
• Reference genome establishment: Applied to improve reference assemblies, exemplified by work on the Chinese hamster genome.
• Assessment of recent assemblies: Used to evaluate and compare newer assemblies, including those analyzed in QUAST-LG studies.

Methodology:

EvalDNA computes multiple assembly-derived metrics and uses a supervised random forest regression model to integrate them into a single quality score; the model explained 86% of the variation in reference-based quality scores on testing datasets.