Documentation needed: contact prediction distance threshold

facebookresearch / esm

Evolutionary Scale Modeling (esm): Pretrained language models for proteins

MIT License

3.16k stars 627 forks source link

Documentation needed: contact prediction distance threshold #94

Closed javierbq closed 3 years ago

javierbq commented 3 years ago

In the jupyter notebook example of [contact-prediction using the MSA transformer] (https://github.com/facebookresearch/esm/blob/0097c27290c7c88c4bc0fc4df3692d2c23a6aa54/examples/contact_prediction.ipynb) there is no mention to the threshold use to define a contact.

tomsercu commented 3 years ago

Hi Javier, thanks for your interest! ~~What exactly is the question / concern? The model outputs continuous valued probabilities per pair of positions. One can threshold at 50% or any other threshold, or the typical contact precision metrics only considers for example the top-L values.~~ EDIT: aha sorry I misread the question/title. It's not about thresholding the probabilities, it's about the distance threshold defining "contact". This is 8A. Will update the notebook, thanks.

tomsercu commented 3 years ago

This is now added as a comment in https://github.com/facebookresearch/esm/blob/master/examples/contact_prediction.ipynb