Association between gene expression levels and clinical symptoms?

Question

I would like to look at association/correlation between gene expression and clinical phenotypes. I am still strugling to know what's the best way to do this. I am pretty new in Bioinformatics but I am craving to learn. I didn't find answer to my questions in the different topics.

To sum up,

• I have RNAseq data from the transcriptome of 10 patients with a disease and 10 healthy donors (CD4+). There are around 1800 differentially genes.
• This disease is really heterogeneous.
• I would like to find genes/ genes cluster among this DE genes, which may be associated/correlated with some symptoms (e.g. eczema, diarrhea) among the different patients.

For continuous variables, I used correlation (spearman to decreas outliers impact). For categorical variable (ordinal or dummies), I tried to do linear regression model with R (dependent variable = eczema, independent variables = genes) but it doesn't work since some genes are highly correlated...

Do you have any suggestions how I can handle this or a good worflow that I can follow?

Thanks

Juliette

Answer 1

To be honest, I think this is an ill-posed problem as you have too few data points while the dimension of the data points is too high.

From a machine learning perspective, we would treat the 1800 gene expressions as a feature vector for each patient. (Of course, you can do some dimension reduction like principal components analysis for the gene expression data) Then you can train a classifier to predict healthy donors/patients for each disease given gene expressions. From a regression perspective, considering you have 1800 genes, I think you can easily find some genes that strongly (albeit suspiciously) correlate with some symptoms.