cicese-biocom
commented
4 months ago Many thanks for your comment about our work!
This framework can be applied for modeling any kind of task other than the antimicrobial peptide classification. However, it is not required to re-train our models. In your case, a new model might be built from scratch using the best settings suggested by us (see train.sh script for an example), or using settings specified by you according to the characteristics of your dataset.
Moreover, we also created a KNIME workflow to automatically built binary classification models by applying 30 feature selectors, 10 individual classifiers, and 4 ensemble classifiers. This framework and results were published at https://onlinelibrary.wiley.com/doi/10.1002/pro.4928.
Best regards,
BIOBRICK
Very impressive work! Congratulations!
Here I wish to ask, wether this framework can be used into other kinds of tasks? Such as enzymatic activity prediction? Because I think here you have presented a universial protein frame without any pre-request. Such kind of network can be used on any kind of protein-related tasks.
For example, I get a dataset containing AA sequences (proteins actually), and have multiple labels for their enzymatic activities for various substance, like H2O2、Sugar or any other things. The protein are classified as 0/1 wether they can catalyze their decomposition. I have tried a lot of solutions and I can't get a good model for my problem. Can I use my dataset to re-train your model and used it on my own task? (with some modifications maybe?)