Noise confounding genes in single cell data

Question

[carolofharvest]

How can I understand if a gene is noise gene in my data? If I suspect a gene is noise gene then how can I be sure about it?

Answer 1

[Michael]

I think you need to revise your question as there is no such thing as a "noise gene" per se. However, the term "noise" comes from signal processing and may mean that that every measurement can be thought of to consist of a "true" value + a random error. For the observer, this results in variation of repeated measurements. In reality, the "true" measurement is inaccessible and approximation is possible due to the law of large numbers and the central limit theorem. Measurement variation is observable for quantitative measurements in biology and any quantitative measurement of a random process. Generally speaking, the more measurements you have, the more confidence you can have both in the measured value as well as the estimation of the amount error.

In conclusion, any gene expression you measure will have a random error attached to it, both from biological as well as technical sources. Every gene is a "noisy gene".

People often filter lowly expressed genes and cells with a low number of expressed genes (possibly an empty bead), low number of reads, or high mitochondrial counts (dead or dying cell) to reduce noise levels in single-cell analyses.