Question related to Fine-Tuning Phase about protein_bert HOT 3 CLOSED

@amalislam675 Yes definitely. The model outputs a vector with the probability of each GO annotation. But again, please notice the issue I linked to.

from protein_bert.

@amalislam675 The concatenation of all hidden layers is implemented in the code by the get_model_with_hidden_layers_as_outputs function (see our demo notebook).

If you're interested in predicting GO annotations, please notice this issue.

from protein_bert.

@amalislam675 The concatenation of all hidden layers is implemented in the code by the get_model_with_hidden_layers_as_outputs function (see our demo notebook).

If you're interested in predicting GO annotations, please notice this issue.

Thanks for your answer.
I have some other questions.
Can we predict multiple labels of GO annotations by giving the protein sequence alone to the model?

For example, total GO terms labels are 5 (GO-term1, GO-term2, GO-term3, GO-term4, GO-term5); protein sequence1 is annotated with GO-term1, GO-term2, GO-term3, while protein sequence 2 is annotated with GO-term 2, and GO-term5.

My aim is to give model the protein sequence1, and protein sequnce2 and ask the model to predict the GO-terms with which both protein sequence 1 and protein sequence 2 are annotated, model should predict the labels for protein sequence 1 are [1 1 1 0 0] and for the protein sequence 2 it should predict [0 1 0 0 1].
Can we predict multilabel outputs from the proteinBERT by giving them protein sequence alone as input and as label [0 0 0 0 0] all zero annotation vector?

from protein_bert.