Hi all,
Just wondering if anyone has come across any algorithms that can impute the similarity of an RNA-seq sample compared with a reference gene expression profile. For example, lets say I have a bulk RNA-seq sample treated with conditionA. I would then like to get an estimated score on how similar it is in it's gene expression profile compared with samples treated with conditionB and conditionC (these two conditions are very different). So what I want is some sort of score, which indicates whether my RNA-seq sample treated with conditionA is more similar to condition B or conditionC.
I have done some research online, and I have the impression that deconvolution algorithms may do the job. But I'm not sure whether algorithms such as CIBERSORTX would give you an estimated score for each sample by comparing with self-defined reference gene expression profiles from conditionB and C. Any insights into this is much appreciated!
I agree with the previous comment. CIBERSORT is dedicated to infer cell composition from a sample based on a gene expression profile. It won't apply to your specific problem. I recommend answers from Comparing similarity of RNAseq datasets and How to evaluate the similarity between two different samples by using RNA-Seq? for your question.
I don't think CIBERSORTx makes sense in this case. I would think a simple pearson correlation is sufficient for measuring similarity in gene expression profiles btwn samples. Or if you have a gene signature for each condition then you can score your samples based on activation or enrichment of the gene signatures.
Thanks for the reply. Pearson correlation is good if I only have a single reference, for example, to see how similar my samples are with only conditionB. But how can I take into account both conditions B and C. I can perform two individual pearson correlations, but from that can I generate a single estimated score to describe sample A's relationship with B and C?