Hello,
I am trying to train a ChromBPNet model on some publicly available single-end bulk ATAC-seq. The pretrained ENCODE bulk ATAC-seq bias model did not successfully remove Tn5 bias (Tn5 motifs present in the nobias model), perhaps not surprising because this pretrained bias model was from paired-end data. However, when I try to train a custom bias model, no matter how I set the bias threshold factor (0.1 to 0.8), the bias model Pearson R in peaks is always < -0.5. I'm wondering if there are any other strategies or pretrained bias models that would make sense to try, and more generally what is your guys' experience with single-end bulk ATAC-seq?
Thanks,
Sahin
Hello,
I am trying to train a ChromBPNet model on some publicly available single-end bulk ATAC-seq. The pretrained ENCODE bulk ATAC-seq bias model did not successfully remove Tn5 bias (Tn5 motifs present in the nobias model), perhaps not surprising because this pretrained bias model was from paired-end data. However, when I try to train a custom bias model, no matter how I set the bias threshold factor (0.1 to 0.8), the bias model Pearson R in peaks is always < -0.5. I'm wondering if there are any other strategies or pretrained bias models that would make sense to try, and more generally what is your guys' experience with single-end bulk ATAC-seq?
Thanks,
Sahin