Figure 2.

Comparison with existing functional scores. (A, B) The performance of our model compared with other functional scores over the validation set (20% of confirmed variants matched with background at 1:10 ratio), as measured by area under the ROC curve (A) And area under precision-recall curve (B). (C, D) The MPRA positive variants from the validation set are embedded along with background using the t-SNE algorithm. (C) We highlight the top 3 percentiles of previously existing functional scores, showing they inhabit different regions of embedding space. (D) We show the location of confirmed MPRA positive variants and the top 3 percentile of our MpraNet score.