0
$\begingroup$

I am trying to align and merge different samples from NCBI. I end up having correlation problem with these sample. The picture below shows an heatmap of the R² by doing a linear regression between 2 samples' count matrices.

picture

Samples A, B and C are from the same project and sample D from another. A, B and C are sequenced using the 10X Genomics for Single Cell V(D)J + 5' Gene Expression, and Chromium Next GEM Single Cell 3ʹ v3.1 for sample D.

All samples have been recovered from NCBI using fastq-dump — internaly threaded to be more efficient — and aligned using cellranger with the default human genome given by 10x.

QC shows some GC content small shifts along side some huge duplication level — even for scRNA-seq.

Am I missing something important when it comes to merge datasets?

Thank you very much

Improve this question
$\endgroup$
3
  • $\begingroup$ Please clarify your specific problem or provide additional details to highlight exactly what you need. As it's currently written, it's hard to tell exactly what you're asking. $\endgroup$
    – Community Bot
    Feb 2, 2022 at 14:20
  • $\begingroup$ Thank you, I am trying to update my comment to get it as clear as possible. $\endgroup$
    – Alexis Finkbeiner
    Feb 2, 2022 at 14:42
  • $\begingroup$ Those correlations don't match your description; A looks very different from the others. Can you confirm that the labels are correct for the graph (e.g. show the code snippet that created it)? I've often found that if create heatmap plots in R with labels defined separately from the matrix, the labels appear in the wrong order. $\endgroup$
    – gringer ♦
    Feb 2, 2022 at 22:09

1 Answer 1

Reset to default
1
$\begingroup$

The picture below shows an heatmap of the R² by doing a linear regression between 2 samples' count matrices.

In addition to doing a heatmap, I'd do a conventional UMAP/t-SNE plot and see how much seperation of the samples there are across all the different clusters/cell types.

Am I missing something important when it comes to merge datasets?

Certainly there are 'technology batch effects' when comparing 3' and 5' data from the same tissue type. You can try to correct for these when merging the dataset by using a batch correction/integration tool (see reviews 1 and 2), but will likely be imperfect.

QC shows some GC content small shifts along side some huge duplication level — even for scRNA-seq.

If you're using a tool like fastQC then note that high duplication levels are expected for (single cell) RNAseq from highly expressed genes.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Hi, I did a umap to better view samples' correlation and it appears that the distance between the samples is proportional to the correlation. As for the technical batch effect, I am looking into it but I find it strange that the three samples A, B and C, which come from the same dataset and use the same sequencing protocol, are so far apart. I will test those method. Thanks you very much for your time $\endgroup$
    – Alexis Finkbeiner
    Feb 3, 2022 at 15:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.