Fig4.b predicted signal

[Witiy]

Hi Jiaqi, It is cool that the predicted signal is highly consistent with other epigenetic signals the model never see. But, as Nvwa is a classifier for gene expression, how to make the model predict such signal ? Are these signal saliency score?

Thanks!

[JiaqiLiZju]

These signal were predictions by scanning sequences along the whole held-out chromosome, instead of saliency score.

The step-by-step procedures were descibed as below: "We extracted 13-kb sequence windows across the whole leave-out chr8 of the human and mouse genomes with 1-kb increments. The ‘bedtools makewindows -w 13000 -s 1000’ command was used to generate these windows, and the ‘bedtools getfasta’ command was used to extract the sequences. The mouse-specific Nvwa model was used for predictions on both the plus and minus strands of the 1-kb-increment window, and the predicted genome-wide signals of each single cell were aggregated to cell-type level. "