Hi @amcerbu
This is a really useful library to work with, thanks! I've been comparing the results from LRR-Annotation to those from LRRpredictor, and I've noticed that some sequences (e.g. 5YUDA, 4KXFB, and 5IRNA) in the LRRpredictor training set hosted in this repo and the original paper have no LRR motifs annotated, despite having canonical LRR domains. I'm aware you didn't generate this original dataset, but given that you used this as the "ground truth" in the paper, I'm curious if you noticed sequences such as these being flagged up? I'd expect your benchmarking of the structural method to actually have a better discrepency value if they weren't excluded :) I was interested in doing some retraining with the original dataset, but worried it might be less "truthy" than it looks (and I'm currently too lazy to manually annotate it)
Thanks again!
Hi @amcerbu
This is a really useful library to work with, thanks! I've been comparing the results from LRR-Annotation to those from LRRpredictor, and I've noticed that some sequences (e.g. 5YUDA, 4KXFB, and 5IRNA) in the LRRpredictor training set hosted in this repo and the original paper have no LRR motifs annotated, despite having canonical LRR domains. I'm aware you didn't generate this original dataset, but given that you used this as the "ground truth" in the paper, I'm curious if you noticed sequences such as these being flagged up? I'd expect your benchmarking of the structural method to actually have a better discrepency value if they weren't excluded :) I was interested in doing some retraining with the original dataset, but worried it might be less "truthy" than it looks (and I'm currently too lazy to manually annotate it)
Thanks again!