A transformer built on DNA language model pretraining, fine-tuned on experimental Z-DNA data.
Attention-based interpretation recovers known Z-forming motifs — alternating purine–pyrimidine repeats, CA/TG and CG dinucleotide tracts — directly from the learned representations, without motif engineering.
Trained on Kouzine et al. 2018.
Described in
Z-flipon variants reveal the many roles of Z-DNA and Z-RNA in health and disease
Trained on