Label assignments for 'Inconclusive' and counterscreen actives

[skearnes]

• Compounds marked 'Inconclusive' in PCBA assays are assigned label -1.
• New Counterscreen class to handle counterscreens: compounds active in both primary assays and at least one counterscreen are assigned label -2.
• parse_pcba_targets.py now takes a --counterscreen argument.

[rbharath]

LGTM for the content of the code.

But, considering our time constraints, I'm going to advocate that we punt on training models with inconclusive compounds and counterscreens for now. There's already a large number of TODOs for the paper, and this one is relatively low priority.

I think that the issue of inconclusive compounds should be properly handled (with empirical measurements of the effects of inconclusive vs. inactive and screen vs. counterscreen) in our MoleculeNet paper.