Hi,
I'm trying to reproduce the paQTL benchmark in Borzoi fig. 6e and am not able to get the same result using the code in the borzoi and borzoi-paper repositories. I think the relevant script is https://github.com/calico/borzoi/blob/main/src/scripts/borzoi_sed_paqtl_cov.py but I'm not sure what settings to use, particularly for --stats. Neither the COVR nor COVR3 setting seems to exactly match the formula described in the paper (Methods, section "Polyadenylation variant effect prediction"). Could you clarify how exactly to reproduce this benchmark?
Hi,
I'm trying to reproduce the paQTL benchmark in Borzoi fig. 6e and am not able to get the same result using the code in the borzoi and borzoi-paper repositories. I think the relevant script is https://github.com/calico/borzoi/blob/main/src/scripts/borzoi_sed_paqtl_cov.py but I'm not sure what settings to use, particularly for --stats. Neither the COVR nor COVR3 setting seems to exactly match the formula described in the paper (Methods, section "Polyadenylation variant effect prediction"). Could you clarify how exactly to reproduce this benchmark?