how to do a reliable single cell sequencing analysis?

Question

[apple4lee]

I want to learn how to do a single cell sequencing analysis, so I watch some video tutorial on YouTube and try to repeat a paper's work. I can get different expression gene of different group in one cluster. However, there are too many parameters that need to be dedicated during the whole analysis. I am a student with no computer science or bioinformatics background, so I am so confused about this. I'm not sure if my choice of parameters is right, I don't have any mathematical or algorithmic knowledge to support my choice. Is there any solution to promise the reliability of my analysis before I already finish learning some related maths lessons?

Answer 1

[ATpoint]

The usual approach for starters would be to follow guided and well-accepted tutorials.

That would (incomplete list) be:

• The Bioconductor single-cell book focusing on Bioconductor tools in R: https://bioconductor.org/books/release/OSCA/

• The Seurat tutorials focusing on the R package Seurat: https://satijalab.org/seurat/

• If you are a Python guy you want ScanPy: https://scanpy.readthedocs.io/en/stable/tutorials.html

• As a general primer I heard good things about https://www.sc-best-practices.org/preamble.html

Be sure to google whatever problems you encounter. Most standard questions have been asked and answered already here or elsewhere.

[apple4lee]

Thank you for your reply. i followed Seurat tutorial of doing analysis, but I found I use different parameters from that paper. This makes me question my whole analysis, cuz I just follow the tutorial and have no idea about why I choose those parameters. For example, I use elbow plot to find out how many PCs to use in downstream analysis, but my numbers are so different from the paper.

[ATpoint]

Without exact code and software it is borderline impossible to fully reproduce a single-cell analysis imo. the linked best practices cover many concepts. Try to learn and then decide whether results make sense. There are many ways to run an analysis that are all correct.