Statement Classification Model

[haohangyan]

This PR implements a statement classification model that predicts whether a statement type is correct based on subject-object pair group information. The features include:

• the evidence count for the statement
• whether either agent belongs to a protein family/complex
• the proportion of the statement evidence count within the total evidence count of the subject-object pair group
• the number of sources supporting the statement
• whether an opposite statement type exists within the subject-object pair group
• the ratio between the evidence count of the statement type and its opposite statement type within the pair group
• the total evidence count within the subject-object pair group

This classification class provides the following functionality:

• predict_pair: predict results from subject and object agent names
• predict_from_rows: predict results from formatted statement rows
• predict_from_hashes: predict results from a list of statement hashes

To implement this feature on db.indra.bio, we also have a large-scale process function that generates a dump file containing all consensus statements for each agent pair. We first generate a pair_to_rows cache file and then predict each pair accordingly. We then create another cache file that we can use when deploying the search page. We parse the information and pass it to the /statement and /search endpoints on db.indra.bio. An example tag looks like Binding-> Phosphorylation ->Activation.