Predicting dynamic expression patterns in budding yeast with a fungal DNA language model

Shorkie, a masked DNA language model fine-tuned on budding yeast RNA-seq, substantially improves gene expression prediction and captures time-varying TF motif contributions across induction courses.

• Why it matters: Predicting gene expression from DNA sequence is difficult because regulatory codes are complex and dynamic, limiting accurate interpretation of noncoding variants and promoter regulation.
• What they did: The model was pretrained on 165 fungal genomes related to budding yeast, then fine-tuned on yeast RNA-seq, including high-resolution transcriptional regulator induction time courses generated in this study.
• The result: Shorkie identifies canonical TF binding motifs, predicts variant effects, improves cis-eQTL classification versus leading sequence-to-expression baselines, and shows high concordance with massively parallel reporter assays while resolving promoter dynamics, splicing signals, and temporal motif usage.

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