Use case from cancer biologist

[allaway]

Issue #3 was getting cluttered so I started a new one for a new use case at @gwaygenomics recommendation. I'm following the format from the first couple of posts there.

Our lab is extremely interested in the maintenance of mitochondria. Some tumor types exhibit greatly increased rates of mitochondrial turnover (i.e. degradation of defective mitochondria, mitophagy). We hypothesize that inhibition of this mechanism could be a tractable therapeutic approach with minimal toxicity to the patient. We have a list of genes that are involved in regulating this pathway but it is technically very challenging and time consuming to identify tumors that exhibit increased mitophagy, even in highly artificial or genetically engineered systems, let alone pathology samples. Of note, mitophagy can utilize a couple of different known pathways (parkin dependent, parkin independent mitophagy).

1. An example of the input list of genes required for and involved in autophagy (not exhaustive, but these are the important ones):
   PARK2
   PINK1
   NIX
   BNIP3
   DRP1
   PARL
2. I'd like to be able to take this classifier and apply it to RNA data from pathology samples, cell lines, PDX models, to determine if they have elevated mitophagy. I'd also like to determine if any particular oncogenic or tumor suppressor pathways are correlated with this phenotype. It would be also nice, but not crucial, to be able to expand the list of genes listed above by manually or automatically including interactors from a database such as Biogrid (example).
3. I'd like to be able to query the quality of the classifier and visualize how my (pathology, etc) samples are classified.

[allaway]

One more potential use case that differs slightly in input/output from the last. Let me know if this is within the scope of this project:

I have a list of genes in which loss or gain of function appears to be highly associated with the formation of glioblastoma. I'd like to generate a classifier that identifies tumors with this mutation profile and then find other tumor types that have this profile. I'd also like to apply this classifier to RNA expression data from my own samples.

1. Input gene list, genes frequently altered in glioblastoma:

NF1 (loss)
TP53 (loss)
SUZ12 (loss)
EED (loss)
SPRED1 (loss)
GLI1 (gain of function, trunc mutation)
EGFR (gain of function, amplification)
IDH1 (wild type = worse prognosis)

2+3. I'd like to be able to take this classifier and apply it to pathological samples for which I have RNA expression data. I'd like to see whether a classifier generated by these data can identify other tumor types outside of glioma. I'd like to determine how survival or time to progression correlates with classifier results.

[gwaybio]

@allaway that sounds awesome, and it will be great to get feedback from our intended audience/user base throughout design.

This type of query is definitely interesting and we could develop some advanced options that could handle it. Even potentially adding in copy number data when the user doesn't necessarily care how gene activity is lost.

We will aim to design cognoma with high modularity so that even if we don't release the service with advanced query support, it can certainly be added in the future. Thanks again for the use case!

[dcgoss]

Input gene list, genes frequently altered in glioblastoma:
NF1 (loss)
TP53 (loss)
SUZ12 (loss)
EED (loss)
SPRED1 (loss)
GLI1 (gain of function, trunc mutation)
EGFR (gain of function, amplification)
IDH1 (wild type = worse prognosis)

http://nbviewer.jupyter.org/urls/cognoma-files.s3.amazonaws.com/media/notebooks/classifier_12.ipynb processed a notebook for glioblastoma with the supplied genes