Addressing two nuisance factors in RNAseq

Question

[jkim]

Hello,

I will be working with approximately 40 bulk RNA-seq samples that include one treatment factor with four levels. However, due to logistical constraints, RNA extraction will be performed on, for example, either Monday or Tuesday, and sequencing library preparation will occur on either Thursday or Friday. Each of these steps will require two days to complete.

I wanted to avoid having the treatment factor confounded by these nuisance factors (RNA extraction day and library prep day), but I’m unsure how best to assign samples across these time points to account for the batch effects (different days for RNA isolation and lib prep). I'm a bit concerned about having a singular model matrix - model matrix is not full rank.

Here is a sample metadata table that mimics the experiment design.

| sample_name | treatment | RNA_extraction_day | library_prep_day | RNA_extract_tech | lib_prep_tech |
| ----------- | --------- | ------------------ | ---------------- | ---------------- | ------------- |
| S1          | con       | A                  | C                | techA            | techB         |
| S2          | con       | A                  | C                | techA            | techB         |
| S3          | con       | B                  | C                | techA            | techB         |
| S4          | con       | B                  | C                | techA            | techB         |
| S5          | stage_1   | A                  | C                | techA            | techB         |
| S6          | stage_1   | A                  | C                | techA            | techB         |
| S7          | stage_1   | B                  | C                | techA            | techB         |
| S8          | stage_1   | B                  | C                | techA            | techB         |
| S9          | stage_1   | B                  | C                | techA            | techB         |
| S10         | stage_2   | A                  | D                | techA            | techB         |
| S11         | stage_2   | A                  | D                | techA            | techB         |
| S12         | stage_2   | A                  | D                | techA            | techB         |
| S13         | stage_2   | B                  | D                | techA            | techB         |
| S14         | stage_2   | B                  | D                | techA            | techB         |
| S15         | stage_2   | B                  | D                | techA            | techB         |
| S16         | stage_3   | A                  | D                | techA            | techB         |
| S17         | stage_3   | A                  | D                | techA            | techB         |
| S18         | stage_3   | A                  | D                | techA            | techB         |
| S19         | stage_3   | B                  | D                | techA            | techB         |
| S20         | stage_3   | B                  | D                | techA            | techB         |

I would appreciate any advice on how to better structure the experiment. These RNA extraction and library preps will be performed by a sequencing core and DNA/RNA extraction core.

Answer 1

[ATpoint]

You have an extraction core? That's awesome.

Anyway, why not keeping it as a single blocking factor. The samples extracted as A get prepped together and those as B also together. That makes a single batch to block for. I don't see the point introducing another batch by splitting the RNA extraction batch during the preps.

[jkim]

Thanks, ATpoint. But I'm not sure if I understand your suggestion. Are you suggesting the following design?

| sample_name | treatment | RNA_extraction_day | library_prep_day | merged_days |
| ----------- | --------- | ------------------ | ---------------- | ----------- |
| S1          | con       | A                  | C                | A_C         |
| S2          | con       | A                  | C                | A_C         |
| S3          | con       | B                  | D                | B_D         |
| S4          | con       | B                  | D                | B_D         |
| S5          | stage_1   | A                  | C                | A_C         |
| S6          | stage_1   | A                  | C                | A_C         |
| S7          | stage_1   | B                  | D                | B_D         |
| S8          | stage_1   | B                  | D                | B_D         |
| S9          | stage_1   | B                  | D                | B_D         |
| S10         | stage_2   | A                  | C                | A_C         |
| S11         | stage_2   | A                  | C                | A_C         |
| S12         | stage_2   | A                  | C                | A_C         |
| S13         | stage_2   | B                  | D                | B_D         |
| S14         | stage_2   | B                  | D                | B_D         |
| S15         | stage_2   | B                  | D                | B_D         |
| S16         | stage_3   | A                  | C                | A_C         |
| S17         | stage_3   | A                  | C                | A_C         |
| S18         | stage_3   | A                  | C                | A_C         |
| S19         | stage_3   | B                  | D                | B_D         |
| S20         | stage_3   | B                  | D                | B_D         |

In my head, the model ~ treatment + merged_days would probably give me a full-rank model matrix :)

[jared.andrews07]

Yes, that is what they're suggesting, as it is the simplest model for what you want (and makes the RNA_extraction_day and library_prep_day factors functionally moot).

[ATpoint]

Yes. As jared.andrews07 says.