目录
目录README.md

Example code of run Tres analysis and reproduce main results

Stage 1: download single-cell datasets

Please CD into the src folder and run “./download.py” first to download all data files, including 31 single-cell training datasets and 10 T-cell transcriptomics validation datasets.

So, you may run “./run.py inx 31” where inx is a number between 0 and 31 to compute the Tres results for each dataset.
We used the NIH high performance cluster (HPC) for parallel computation as “./hpc_submit.py 31”. You may re-write this file for your local HPC.

The output will be available in data/output, named with the cancer type followed with database accession ID.

Stage 2: merge computation scores into signature files

After computing the correlation results for each dataset, you can run “./run.py” to generate the merged Tres signature files in data/output:
1, merge.TGFB1, merge.TRAIL, merge.PGE2: Tres scores among all samples for each dataset.
2, merge.Median: Median signatures among all immunosuppressive signals enumerated above.
3, merge.Median.AUC: Area under the ROC curve to measure the quality signatures in “merge.Median”, using T-cell persistance markers (Krishna et al., Science 2021) as the evaluation standard.
4, merge.signature: One overall Tres signature, which is the median across all datasets.

Stage 3: predict clinical response of immune checkpoint inhibitor (ICI), CAR T, and adoptive cell transfer

After creating the Tres signature, you can run “./predict.py” to evaluate qualities of therapy response predictions (Figure 2 in the manuscript). All results files are named as *.pdf in data/output.

Task 1: predict responder versus non-responder status using T cells from pretreatment tumors for ICI or infusion products for CAR T:
This task will analyze cohorts:
Atezolizumab+Paclitaxel_Pre_Zhang2021_TNBC
CD19CAR_Infusion_Fraietta2018_CLL
ICB_Post_SadeFeldman2018_Melanoma
ICB_Pre_SadeFeldman2018_Melanoma
Nivolumab_Post_Caushi2021_NSCLC
anti-PD1_Pre_Yost2019_BCC
The clinical outcomes are responders or non-responders. Thus, we can generate Receiver Operating Characteristic (ROC) curves to compare signatures on predicting therapy response.
The result for each dataset will contains one boxplot (.boxplot.pdf) and one ROC curve (.ROC.pdf).

Task 2: predict cell therapy survival outcome using data from pre-manufacture samples for CAR T or adoptive cell transfer (ACT):
This task will analyze cohorts:
CD19CAR_Pre-manufacture_Chen2021_Bcell
ACT_Pre-expansion_Lauss2017_Melanoma
The clinical outcomes are survival durations and the sample expression profiles were taken before manufacturing the therapeutic T cells.
Thus, we can analyze whether correlation with the overall Tres signature can predict the survival outcome through Kaplan-Meier plots and Cox-PH regression.
The result for each dataset will contains one Kaplan-Meier plot (.kmplot.pdf) and one barplot (.bar.pdf) presenting Wald test risk z-scores.

关于
1011.0 KB
邀请码
    Gitlink(确实开源)
  • 加入我们
  • 官网邮箱:gitlink@ccf.org.cn
  • QQ群
  • QQ群
  • 公众号
  • 公众号

©Copyright 2023 CCF 开源发展委员会
Powered by Trustie& IntelliDE 京ICP备13000930号