The abstract is read at the publisher; the summary is JClub's.
Self-supervised pretraining of a Genomic Pretrained Network on 455 apicomplexan genomes improves regulatory signal prediction in Plasmodium falciparum, especially with fewer than 3,000 labeled examples.
- Why it matters: Foundation-model gains are established in large mammalian datasets, but their value for small, data-sparse genomes like malaria parasites has not been directly tested.
- What they did: The model was trained by masked-nucleotide reconstruction across 455 genome assemblies spanning 76 apicomplexan species, then fine-tuned on a held-out P. falciparum chromosome to predict transcription factor binding, chromatin accessibility, histone marks, and 10 chromatin states.
- The result: Pretraining plus fine-tuning improved prediction for most epigenomic features, with the strongest benefit in datasets under 3,000 training examples, while a human-pretrained model transferred poorly.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.