Thanks for all your amazing work! I was looking through the PRISM drug screen data and saw some cases where the same drug had multiple broad IDs (e.g. erythromycin maps to BRD-K63550407-001-12-7 and BRD-K63550407-001-13-5 in primary-screen-replicate-treatment-info). More confusing, when I look at the effects output in primary-screen-replicate-collapsed-logfold-change, I see nearly no correlation between the response vectors in these two experiments. Are they different drugs somehow? Was the same drug used twice and the lack of correlation just speaks to high levels of noise in the data? Is there something else going on? Thanks so much!
The difference in the BRD id’s is basically referring to different batches of the compound used in those screens. Theoretically, this shouldn’t change the result much, but unfortunately, they occasionally do on such large screens. I would say it is probably safe, the one that seems like the inert one.
That has been said, as you pointed out, the noise level also relatively high in this dataset, especially if there is not a strong responder. This seems like the case in your example; in particular, neither of the profiles seems like killing more than 50% on any of the cell lines in our cell line panel.