Update survival tp53/telomerase analysis

jharenza commented 2 years ago

Purpose/implementation Section

What scientific question is your analysis addressing?

This is a followup to https://github.com/AlexsLemonade/OpenPBTA-analysis/pull/1208 and https://github.com/AlexsLemonade/OpenPBTA-analysis/pull/1212

What was your approach?

Removed samples with missing survival data to get accurate numbers on plots
Added cancer group to univariate analysis and removed plots, which were not super informative
Initially, the multivariate analysis did not consider interactions between covariates, so I added them here.
Added a within cancer group assessment of tp53 scores and telomerase scores, rather than within HGAT or non-HGAT
Renamed the notebook to describe the contents more accurately
Did not add README since #1257 is in queue

What GitHub issue does your pull request address?

NA

Directions for reviewers. Tell potential reviewers what kind of feedback you are soliciting.

Which areas should receive a particularly close look?

I really only updated current code to try to figure out the best way to look at these data and did not do any optimization.

I think that we can include the coxph plot from chunk 7 in Figure 4, and include some of the tabular statistics for interactions/cancer group tp53/telomerase results/interactions within the text.

Perhaps optimization and/or crisp figure generation is something @runjin326 or @sjspielman can work on? Note: @runjin326 is working on optimizing #1264, so it is possible this analysis can piggyback onto the functions being generated there.

Is there anything that you want to discuss further?

Does this workflow make sense?

Is the analysis in a mature enough form that the resulting figure(s) and/or table(s) are ready for review?

yes

Results

What types of results are included (e.g., table, figure)?

tables, KM curve, hazard plots

What is your summary of the results?

Univariate analysis

tp53 scores and telomerase scores are a significant hazard in that higher scores of either are independently associated with poor prognosis (OS)
HGAT group is also significantly associated with poor prognosis
EPN, LGG, MB, CRANIO, neurofibroma plexiform, and schwannoma have significant survival benefit (as evidenced by negative estimates and as expected) and DMG has a significant survival hazard.

Multivariate analysis

Higher tp53 scores and higher telomerase scores are each associated with a significant survival hazard (positive estimates) and there is a significant interaction between the two with lower values (negative estimate) having a survival benefit. I am not 100% sure what a negative estimate here means - does it mean both scores are negative or one of the two are negative (which one).
Here, there is no significant difference between non-HGAT and HGAT (not 100% sure why, but looks like there is too much variance in the HGAT group and it seems like with multiple covariates, this is just not significant).

Within cancer groups:

Only MB and HGG had significant results.
For MB - low tp53 and low telomerase scores had a significant survival benefit, and there is a significant interaction between the two. However, I am also not sure how to interpret the positive estimate of this interaction.
For HGG - higher tp53 and higher telomerase scores were a significant hazard. There is also an interaction, but this has a negative estimate.

Reproducibility Checklist

[x] The dependencies required to run the code in this pull request have been added to the project Dockerfile.
[x] This analysis has been added to continuous integration.

Documentation Checklist

[ ] This analysis module has a README and it is up to date.
[ ] This analysis is recorded in the table in analyses/README.md and the entry is up to date.
[x] The analytical code is documented and contains comments.

jaclyn-taroni commented 2 years ago

Just adding a link to this comment where interaction was discussed because I keep looking for it and the title for the PR it's on doesn't have a title that makes me think it would be there https://github.com/AlexsLemonade/OpenPBTA-analysis/pull/1216#issuecomment-1011153246

sjspielman commented 2 years ago

@jharenza I will make the publication-ready figures for all the survival notebooks once we have the modules solidified. We can open separate issue(s) there once we have a better sense which survival analysis/es will be included in the manuscript.

In terms of optimizing the code, I'll start with this PR. We might want to have this PR block #1264 depending on what kind of reorganization has to happen here that would be shared with the #1264. After I review this a bit will have a better sense.

sjspielman commented 2 years ago

@jharenza Is there a specific reason the template functions shown in survival-analysis_template.Rmd aren't used? The main difference I can see is that this code used survival years instead of days, but this is easy to get around - is there a specific reason we are using years here instead of days? The best way to approach this updated survival notebook and that coming up in #1264 is to use the existing functions and template as much as possible to ensure overall modularity throughout the analysis.

I'm going to file a PR to your branch that leverages the util/survival_models.R functions and we can go from there!

jharenza commented 2 years ago

Thanks @sjspielman. I think years were used because the x-axis was too unreadable using days, but that can probably also be fixed using fixed spacing in the plot. I'm not entirely sure about the functions in general- could be that those functions were univariate only and I have a vague memory that there might have been some plot rendering issues.

jharenza commented 2 years ago

Just adding a link to this comment where interaction was discussed because I keep looking for it and the title for the PR it's on doesn't have a title that makes me think it would be there https://github.com/AlexsLemonade/OpenPBTA-analysis/pull/1216#issuecomment-1011153246

Thank you for finding that. I knew we discussed but I didn't remember why we didn't look at this. I think that given the question we have - which is do the TP53 and telomerase scores have some combined predictive value (does adding the p53 score change the telomerase prediction) - I think we need to add the interaction terms.

I still need help with is interpretation, though.

sjspielman commented 2 years ago

I still need help with is interpretation, though.

Interpreting these interaction models is not going to be particularly straight-forward, beyond just plainly stating which interactions are significant and what coefficients are. This is not something is easily described beyond pure mathematical terms; I don't have "gut sense" beyond that, as it were.

jharenza commented 2 years ago

Thanks for the updates @sjspielman - I will take a look in more detail tomorrow, but at first glance, I noticed that in the "TP53 score * Telomerase score, for each cancer group." section, the analyses are not subsetting each group (at least I do not think they are doing it properly). The N being plotted is the same for every group (all 548 samples) and all tables and plots are identical. I had this issue before, but will have to go back and check tomorrow how I had to handle it. Survival analysis is pesky!

sjspielman commented 2 years ago

@jaclyn-taroni @jharenza This is now ready for a look. I'm pretty satisfied with this notebook at this stage! Some of the changes I included (in particular a function) can be used in the immune survival notebook I'll get to next, as well.

Summary of the changes I've made:

(quick jump to rendered notebook)

There is now a single function added to util/survival_models.R that will handle model fitting and exporting as RDS objects, for both univariate and multivariate. As stated, this function can be used for the immune survival notebook coming up next.
No longer read in polya telomerase scores since they are removed during processing anyways
Added TOC for improved file navigation
For the long rank cancer groups model, I added pairwise comparisons among cancer groups, corrected with Bonferroni,+ + I removed plots in favor of printing result tables and exporting fill model fits as RDS. Plots were very busy and not always legible, so I suggest we make plots separately for the manuscript without necessarily making plots in this module.

Note also that the data prep is actually pretty different between the immune scores survival analysis and this one, so we do not need to have a shared data prep script.

jharenza commented 2 years ago

this looks good to me now - ready for re-review @jaclyn-taroni !

sjspielman commented 2 years ago

This has passed CI successfully before merging master in, so this PR be merged in before checks are complete.

AlexsLemonade / OpenPBTA-analysis