search

alexllc / HTE

0 stars 0 forks source link

Other tx variables Immune proportion #3

Closed alexllc closed 3 years ago

alexllc commented 3 years ago

November 9, 11:51 AM

Using BRCA and HNSC, CIBERSORT proportion, three types of aggregate (lm06, lm11, lm22, detailed can be found in /home/yujia/Project/2020-11-06-TCGA_pancancer_immune/dat/cibersort_prop/cluster.md) None of the cell type tx can pass the test (i.e the pval of tau.risk (fixed YW) by permutation covariates < 0.05).

Using BRCA and HNSC, EPIC proportion, ran by Alex Only one cell type tx (CD8 T-cell) in HNSC can pass the test (i.e the pval of tau.risk (fixed YW) by permutation covariates < 0.05).

Should we include the immune proportion result?

Dear Yujia,

Thanks a lot and will get back to you soon. If no results pass permutation test, you can probably still pick one or two (eg those with the most sig split-half correlation with sig. simes p) to do Xgboost. Otherwise, can perform some further xgboost modeling on top genes from expression/mutation from Alex's side. Kai's results will be available later. Thanks.

Best regards, Hon-Cheong

On Mon, Nov 9, 2020 at 2:37 PM SHI, Yujia yj.shi@link.cuhk.edu.hk wrote: Dear Dr.So,

I have already finished the analysis using Alex HTE code with CIBERSORT immune cell proportions (three aggregates, lm06, lm11, lm22) on HNSC and BRCA dataset. Unfortunately, no cell type tx can pass the permutation test and enter the XGboost learning step. For Alex's result on BRCA and HNSC using EPIC proportion, only one cell type (CD8 T cell) in HNSC can pass the permutation test. Some details can be found on the google doc "ideas for finishing HTE (Nov 2020) ". We may discuss about the progress and direction of immune cell analysis later this week. Bayesian tuning code and XGboost learning code for immune cell analysis is finished as well, thus it won't take a long time for the rest part (less than 1d). We may focus on the HTE analysis first.

Analysis code: 234: /home/yujia/Project/2020-11-06-TCGA_pancancer_immune/bin/

Analysis log: 234: /home/yujia/Project/2020-11-06-TCGA_pancancer_immune/log/

Analysis result: 234: /home/yujia/Project/2020-11-06-TCGA_pancancer_immune/res/

Best Regards, Yujia

alexllc commented 3 years ago

Cell type (cibersort 22 cell types -> now we aggregate them to 6 cell types 1st), (we will look at the distribution to determine the cut-off and whether to run tx as cont. var) Set a rough rule on the proportion of subjects having 0% of the cell type to consider 0% as a category

alexllc commented 3 years ago

run.hte no longer processes treatment group assignment steps, W vector is used as a separate input in the form of a W_matrix with row patient entries and column treatment variable entries. An additional function create_tx_matrix for creating the W_matrix has been created in the TCGA_data_function.R script. See commit 15216b1 for details.