Experiment Design - New Drugs

[gwaybio]

@mekelley and @ayberman are getting ready to ramp up data collection. Specifically, we will collect Cell Painting data from 7 resistant clones and a wildtype parental line undergoing two drug treatments and a DMSO negative control. The drugs are ixazomib (proteasome inhibitor) and CB-5083 (p97 inhibitor). The cells will be treated for four hours.

Clarification Questions

• Will we also collect Cell Painting profiles with different drug doses?
• Will the 7 resistant clones be selected independently per drug? In other words, will there actually be 14 new resistant clones?
• Do we need to include a wild-type clonal line?

Wildtype Clonal Line Discussion

In my opinion, we definitely should try to mirror the same clonal selection procedure for the wildtype parental line to acquire a wildtype clonal line. Our approach in identifying morphology features of resistance mechanisms will be skewed by clonal selection signal. In other words, we're likely to obscure the resistance signals by also isolating the clonal selection signals.

I view the wild-type parental line as a really great validation resource. We should see a lot of heterogeneity in these samples, and could validate some resistance signatures we find by comparing wildtype clones and other resistant clones.

If we use the same clonal selection procedure for selecting resistant cells to the new drugs, then I think using the same WT clones for batches 4-7 (see #40) is sufficient.

[ayberman]

To answer @gwaygenomics clarification questions:

Clarification Questions

• Will we also collect Cell Painting profiles with different drug doses? We were not planning on doing so.The logic behind doing this with clones A and E was that we knew that Clones A and E harbor mutations in PSMB5, among other mutations, so we hypothesized that the bortezomib resistnce signature they generated would be linked to proteasome perturbation. For the other bortezomib and other drug clones, we have no genetic sequencing, so it would be difficult to understand the cell painting signatures they generate.

• Will the 7 resistant clones be selected independently per drug? In other words, will there actually be 14 new resistant clones? Yes, there are 14 independent new clones. 7 are Ixzomib resistant, and 7 are CB-5083.

• Do we need to include a wild-type clonal line? To clarify, do we have enough data from the WT clones already processed, or should we include the WT clones in this new experiment so that their signatures can be more directly compared to the new clones (ie, is the data from WT and drug "tolerant" clones only comparable if they were collected on the same plate on the same day)?

[gwaybio]

Yes, there are 14 independent new clones. 7 are Ixzomib resistant, and 7 are CB-5083.

cool 😎

so we hypothesized that the bortezomib resistnce signature they generated would be linked to proteasome perturbation. For the other bortezomib and other drug clones, we have no genetic sequencing, so it would be difficult to understand the cell painting signatures they generate.

I think I am missing the connection of drug dose to sequencing... can you clarify? Also, if our observations are leaning us towards the identification of a "drug tolerance" signature, then dose information is how we support it.

Although this question is directly related to the scope and target of the journal. Here is how I see the paper:

Paper with dose for one drug (bortezomib) and without dose for two drugs

1. Show drug tolerance signature in single cell bortezomib-resistant clones
   • Benchmark that this signature is robust to across-plate, across-batch, and across-clone comparisons
2. A deep dive into the Cell Painting features that this model learns
   • What do these features tell us about bortezomib tolerance?
3. Does the signature also predict tolerance mechanisms to other drugs - this will test how specific the signature is.
   • Depending on the answer, can we build more specific signatures?
4. [just an idea] given the tolerance prediction heterogeneity we observe in the single cell wild-type parental line, can we predict which treatment will result in lower cell survival?

Paper with dose for three drugs

1. Same as 1 above
2. Same as 2 above
3. Confirmation that we observe tolerance signatures in two additional drugs
4. How much information is shared between the tolerance-signatures? How specific are they? Do they share commonalities? Can we parse the feature differences?
5. Same as 4 above, with an added dimension of predicting similar viability with different drug dose

Summary

I think we have a nice story either way we go. I think including dose information might be a bit more compelling, but, again, only if it does not add too much additional work.

To clarify, do we have enough data from the WT clones already processed, or should we include the WT clones in this new experiment so that their signatures can be more directly compared to the new clones (ie, is the data from WT and drug "tolerant" clones only comparable if they were collected on the same plate on the same day)?

Gotcha! I thought that the original question was, "should we also include WT clones on this plate that we already made previously". The answer to that is yes, please! Is that the question?

Or, is the question, "should we perform a new round of selection to get WT clones, grow them up, and then add to new plates"? The answer to that is, yes, please, if it is not too much trouble and would delay us too much.

[mekelley]

Thanks for the feedback @gwaygenomics! If we were to generate new wild type clones, then @ayberman and I would be delayed several weeks.

Instead, could we use 4 wild type clones that we generated previously (batch 3) but haven't been thoroughly characterized yet (not used in batches 4-7)?

[gwaybio]

Thanks for this note!

Instead, could we use 4 wild type clones that we generated previously (batch 3) but haven't been thoroughly characterized yet (not used in batches 4-7)?

I think that is perfectly reasonable 💯