search

greenelab / snorkeling

Extracting biomedical relationships from literature with Snorkel 🏊
Other
58 stars 17 forks source link

Quick fix #90

Closed danich1 closed 4 years ago

danich1 commented 4 years ago

This PR updates the plots for my manuscript. Should be a quick review.

danich1 commented 4 years ago

  1. Edges already existing in Hetionet are considered the true positives. No that is too strong of a statement. It is better to say that our model has more uncertainty when predicting gene interacts gene edges.

  2. For the sake of PSB manuscript leaving the caption in is the best option. Ultimately, I can go either way on this.

  3. CoCoScore is actually the published baseline model's name. Cant remove or change it. "Our Model" was suppose to be changed, but I missed making that change before submission. Also, these methods are doing very different things to classify edges.

Our Model -> predict edges by classifying sentences aka supervised machine learning. CoCoScore -> a hybrid approach using co-occurrence scores and logistic regression.