[Question]: Why not include cell type detection in Hunflair?

[skywalker2202]

Question

First of all, thank you for this resource. This is more of a curiosity. CRAFT v4 contains cell type information as well. Did you guys try for Named Entity Detection in Hunflair? If yes, how did it work out?

[alanakbik]

Hello @skywalker2202, the hunflair models do predict entities of type "cell line", but I am not sure if this is what you are looking for. Probably @mariosaenger can best answer this.

[skywalker2202]

Hey there. I meant cell types and not cell line information. For instance in a sentence like :
['Number' 'of' 'glucocorticoid' 'receptors' 'in' 'lymphocytes' 'and'
'their' 'sensitivity' 'to' 'hormone' 'action' '.'] -> the lymphocytes are a cell type.

[mariosaenger]

Hi @skywalker2202!

Sorry for the late response. For HunFlair we have focussed on the basic entity types that a) are also supported in many other NER taggers and b) for which there are several gold standard corpora (as training on these is a key methodological contribution of HunFlair). There was on special reason to exclude cell types annotation from CRAFT.

Have you tried to train a CRAFT-based model for cell types on your computer, yet? I believe it should work right away. Refer to https://flairnlp.github.io/docs/tutorial-training/how-to-train-sequence-tagger for instructions how to train a NER model in Flair.

If you have any questions or issues while training the model, please let me know.

Best,
Mario