Could not reproduce the results #2
Description
Hi Dr. Wang,
Sorry for submitting the issue to the wrong repo. I actually was using the script and model weights in this repo to reproduce your results in the NAR paper. The question remains the same: "I was trying to reproduce the result in your paper with the trained model weights in this repo. To construct the benchmark dataset, I followed your method in the NAR publication (Wang, D. et al., Nucleic Acids Research 48, W140–W146, 2020) by using the modified sites of proteins released after 2010 as positive data and the rest sites in the same protein as negative data. However, the model performance was not as good as your publication. Especially in the o-glycosylation, ubiquitination task, and phosphoserine/threonine tasks. Could you give more details about how you constructed the independent benchmark and how to produce the results in Table 1?" Thank you so much!
Best,
Thomas