sorAI - a suite of machine learning models to predict plant polyploidy from target capture data

sorAI predicts diploid versus polyploid genomes in ferns from GoFlag target capture data, with a random forest achieving ~71–78% accuracy across test datasets and sequencing depths.

• Why it matters: Ploidy is hard to measure for individual land plants, and existing frequency-based inference methods underperform as target-capture datasets grow.
• What they did: The authors trained a CNN per GoFlag probe locus and built ensemble models using two loci plus a random forest using all loci, then evaluated them on held-out fern datasets and an independent Lygodium test set.
• The result: sorAI outperforms nQuire and nQuack on held-out GoFlag data, leverages positional and covariation signals beyond allele frequencies, and maintains performance as sequence depth decreases.

AI summary of the abstract · may contain errors · not medical advice · read the paper.