search

openvax / vaxrank

Ranked vaccine peptides for personalized cancer immunotherapy
Apache License 2.0
54 stars 22 forks source link

Make use of clonality information #4

Open hammer opened 8 years ago

hammer commented 8 years ago

Nina really wants clonality information to be used in our pipeline!

ihodes commented 8 years ago

@iskandr

Cool; I remember hearing about this, but don't remember seeing this in the IND anywhere. I'm probably due for a re-read; do you (@iskandr) have the newest version of the IND?

iskandr commented 8 years ago

It's not in the study protocol for PGV, though since no patients have enrolled yet we could theoretically modify the protocol. However, the last time I checked we still hadn't found a reliable method for estimating clonality from WES -- each tool will give us discordant numbers for a variant. @armish is that right?

Let's say we did have a good way of estimating clonality from WES. Do you all have thoughts on how it should be integrated into variant ranking? Currently vaxrank multiplies the sum of normalized binding scores and sqrt(n_alt_reads) to combine expression and MHC binding for each vaccine peptide. We can throw in clonality as well, but it's already a somewhat arbitrary formula. Should we try to optimize the ranking in some way? And if so, what data is available to optimize against?

JPFinnigan commented 8 years ago

Since we already use of allele-specific expression estimates in scoring, doesn't incorporating clonality into the scoring seem a little redundant? Would you want clonality estimates based on VAF from the WES, to supplant ASE from the RNASeq?

Nevertheless, it would be interesting to pilot this with the B16 data. Before considering this for PGV it seems like it would be worthwhile to know how the WES-based tools perform in a realistic but controlled setting--both the B16.F1 and B16.F10 lines were single-cell sorted prior to WES so they are fully clonal and lack stromal/vascular/leukocytic contamination, but they exhibit strong signs of copy-number variation. Could be neaT, but I've never been able to get the tools working on my laptop.

iskandr commented 8 years ago

@armish The B16 samples do provide an interesting test case. I suspect that every clonality estimator will be thrown off by the complex copy variants.

iskandr commented 8 years ago

@JPFinnigan I remember us talking about this before and it does seem potentially redundant. However, if a minor clone is making a lot of some mutant transcript than its more abundant siblings (which lack that variant) then clonality could potentially be useful (though that added benefit is still highly speculative).

JPFinnigan commented 8 years ago

However, if a minor clone is making a lot of some mutant transcript than its more abundant siblings (which lack that variant) then clonality could potentially be useful (though that added benefit is still highly speculative).

Perhaps, but I'm not sure how one would disambiguate the scenario you're describing from one of uniformly high expression of a variant w/in the total population given that we have bulk WES/RNASeq as opposed to multi-regional sequencing. I think it's going to be a really hard problem to address given the limitations of the data we're going to collect.

That being said it would be really neat, since we're already thinking about the issue of clonal selection and therapeutic resistance, if we could work this into the correlative/exploratory aspect of the first PGV study. For example, if we have the material for some patients, perhaps we could save isolated DNA/RNA from multiple regions of a tumor to revisit if/when a PGV patients relapses. IMO, the outgrowth of antigen/MHC/B2M-loss variants would be the strongest evidence of efficacy we could gather in a Phase I trial.