ERCC spike-in is a set of synthetic controls developed for RNA-Seq. I'm interested in using it to normalize my RNA-Seq samples. In particular, I'd like to use the spike-ins to remove technical bias and any variation that should not be part of my analysis.
The site doesn't give any details on how I can do that.
Q: What are the possible normalization strategies? Can you briefly describe them?
You may consider using RUVSeq. Here is an excerpt from the 2013 Nature Biotechnology publication:
We evaluate the performance of the External RNA Control Consortium (ERCC) spike-in controls and investigate the possibility of using them directly for normalization. We show that the spike-ins are not reliable enough to be used in standard global-scaling or regression-based normalization procedures. We propose a normalization strategy, called remove unwanted variation (RUV), that adjusts for nuisance technical effects by performing factor analysis on suitable sets of control genes (e.g., ERCC spike-ins) or samples (e.g., replicate libraries).
RUVSeq essentially fits a generalized linear model (GLM) to the expression data, where your expression matrix $Y$ is a $m$ by $n$ matrix, where $m$ is the number of samples and $n$ the number of genes. The model boils down to
$Y = X*\beta + Z*\gamma + W*\alpha + \epsilon$
where $X$ describes the conditions of interest (e.g., treatment vs. control), $Z$ describes observed covariates (e.g., gender) and $W$ describes unobserved covariates (e.g., batch, temperature, lab). $\beta$, $\gamma$ and $\alpha$ are parameter matrices which record the contribution of $X$, $Z$ and $W$, and $\epsilon$ is random noise. For subset of carefully selected genes (e.g., ERCC spike-ins, housekeeping genes, or technical replicates) we can assume that $X$ and $Z$ are zero, and find $W$ - the "unwanted variation" in your sample.
We have added ERCC spike-ins to all our RNASeq data, just in case other people might find it useful in the future. However, I have never used it in my own analyses because I can't think of a reasonable way that it could be used.
The typical recommendation for ERCC is to add it in proportion to the input RNA amount, but that makes an assumption that total cell RNA counts are similar across different cells (which is demonstrably false by looking at single cell RNASeq results).
I have yet to think of a situation where ERCC would provide better results than a "housekeeping" gene set sampled from the original reads.
