Batch correction and normalization for multi-sample scRNA-seq analysis

[liyapingdoct]

Hi SecAct team,
I would like to infer secreted-protein signaling activity for the same cell type/state and biological condition, but the cells originate from multiple samples and independent cohorts.
What is the recommended workflow for handling sample- and cohort-specific effects? Specifically:

1. Should expression be pseudobulked separately for each sample × cell type before running SecAct?
2. Should batch correction be applied before activity inference? If so, which expression scale or method is recommended, and should integrated/Harmony-corrected values be avoided?
3. For cross-cohort comparisons, is it preferable to infer activity per sample and include cohort/batch as a covariate in the downstream statistical model rather than pooling all cells?
   An example workflow for multi-sample or multi-cohort scRNA-seq data would be greatly appreciated.

[data2intelligence]

Hi Liyaping, Please see my suggestions in the lines below: On Sep 14, 2026, at 11:27 PM, liyapingdoct ***@***.***> wrote: [https://avatars.githubusercontent.com/u/138564818?s=20&v=4]liyapingdoct created an issue (data2intelligence/SecAct#8)<#8> Hi SecAct team, I would like to infer secreted-protein signaling activity for the same cell type/state and biological condition, but the cells originate from multiple samples and independent cohorts. What is the recommended workflow for handling sample- and cohort-specific effects? Specifically: 1. Should expression be pseudobulked separately for each sample × cell type before running SecAct? Yes, please pseudobulk separately for each sample x cell type. 1. Should batch correction be applied before activity inference? If so, which expression scale or method is recommended, and should integrated/Harmony-corrected values be avoided? I would suggest you trying two ways: 1) just run analysis on the raw data with NO batch correction; 2) run Harmony before all downstream analysis. Then, you can compare the results and see which versions make better sense to you. Overall, the decision to use batch effect or not really depends on your specific dataset and scientific questions. Trying two ways and evaluating your result is always a good start. 1. For cross-cohort comparisons, is it preferable to infer activity per sample and include cohort/batch as a covariate in the downstream statistical model rather than pooling all cells? An example workflow for multi-sample or multi-cohort scRNA-seq data would be greatly appreciated. Yes, please infer activities per sample and include cohort and batch as covariates in downstream statistical models. For all of our multi-sample analysis in the paper, the code are available at the “Code Availability” section as https://github.com/data2intelligence/SecAct_main Hope these points can help you. I also copied Beibei and Seongyong here, in case they have further suggestions. Peng Reply to this email directly, view it on GitHub<#8?email_source=notifications&email_token=AQRREF5HWLSUS5J6YILGSA35PCZK7A5CNFSL4Z3JMQ5C6L3HNF2C22DVMIXUS43TOVSS6NJUGU3TCMJZGI4DDJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KYZTPN52GK4S7MNWGSY3L>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AQRREF5QFDO2JQF4YSP2T535PCZK7AVCNFSNUABFKJSXA33TNF2G64TZHM3TENJRHEYDSOBVHNEXG43VMU5TKNBVG4YTCOJSHAY2C5QC>. Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS<https://github.com/notifications/mobile/ios/AQRREFZYKCJ62JHOZRFXMN35PCZK7A5CNFSL4Z3JMQ5C6L3HNF2C22DVMIXUS43TOVSS6NJUGU3TCMJZGI4DDJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2KUZTPN52GK4S7NFXXG> and Android<https://github.com/notifications/mobile/android/AQRREF5HE5GUF2XYBFHZJO35PCZK7A5CNFSL4Z3JMQ5C6L3HNF2C22DVMIXUS43TOVSS6NJUGU3TCMJZGI4DDJTSMVQXG33OVJZXKYTTMNZGSYTFMSSWK5TFNZ2K4ZTPN52GK4S7MFXGI4TPNFSA>. Download it today! You are receiving this because you are subscribed to this thread.Message ID: ***@***.***> CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you recognize the sender and are confident the content is safe.

[beibeiru]

Just one additional point:

For each run of SecAct activity inference, the inferred activity represents relative values across all input samples within that run. Therefore, results from different runs cannot be directly compared.

Best,
Beibei