How to top n values from .predict_proba in multioutputclassifier?

I am following MultiOutputClassifier technique to predict roles (the data are transformed to numeric so that's not a concern)

I want to use .predict_proba() and extract top 4 values along with its classes.(there are 5 output variables role0, role1, role2, role3, role4, role5).

I am not sure how to do this.

While doing normal analysis we use this to extract top n values for each observation.

proba_k=k.predict_proba(xval)
best_n=np.argsort(-proba_k, axis=1)[:, :4]

My concern is to predict top 4 roles in role0.

           role0                             role1                   role2                role3          role4                     
9469    Small Business Owner Operator   Operations Director        Chief Executive Officer  Service Manager             Manager Commercial Lines
1433    Internal Sales Coordinator      Financial Systems Analyst  Corporate Paralegal      General Ledger Accountant   Project Accountant

I am sorry if the table looks a bit clumsy.