Gene Expression Timeline Analysis for N=1

Question

[ATRX]

Hi,

I have a timeline RNA-seq dataset across four different time points but only for one patient. These four different time points can be annotated as untreated and then three different treatments. I used STAR for the alignment and calculated TPM values using RSEM for further analysis.

I have a couple of questions:

1. What is the best way to find the genes apart from using variance/sd/z-score as a metric?
2. Is there a way to test these top genes that may be differentially expressed in other diseases or disease-treated settings? Or, something similar to GO or pathway analysis?

Disclaimer: I want to apologize for this post, as this may annoy a lot of statisticians and bioinformaticians. I totally understand that my analysis will be limited, and not much can be interpreted from it. Still, my collaborator, a physician-scientist, is interested in the analysis and totally understands the limitations of the study. We want to pursue it because this is a patient sample, and getting a vast number of patients with the same diagnosis/disease/mutation and then applying similar treatment is not easy.

Answer 1

[ATpoint]

You say time series but it's actually discrete treatments, so no time series. It is an unreplicated within patient. I would probably apply the suggestion in the edgeR user guide in section 2.12 (essentially faking replication for the sake of getting some results), and from that take the top genes by PValue for gene ontology or manual inspection. Or take the top like 500 genes and subject to heatmap and clustering, maybe some patterns emerge that are worth following up. Or do dimensionality reduction like PCA or NMS and see whether the eigengenes suggest interesting separation.

Unreplicated differential analysis has been asked many times before, please search for these many posts.