Fig. 3: The distribution of prediction correlation between predicted and actual gene expression values.
From: Benchmarking DNA foundation models for genomic and genetic tasks
The histograms showing the prediction correlation on all genes using (a). HyenaDNA and (b). HyenaDNA-450 K which takes longer input sequences (196 K bps). Both models demonstrate similar distribution patterns with a slight positive skew in the histogram, indicating that while most genes have modest predictability, a subset of genes shows stronger sequence-expression relationships.
