1
$\begingroup$

I'm working on a gene expression dataset from patients who either have systemic sclerosis or not. I normalized the data with housekeeping gene values and scaled the expression values for each DNA probe results to a scale from 1 to 10. I then calculated the ANOVA F-scores for each probe to find the ones that are likely to be better predictors of disease. I did a PCA analysis using the top 135 probes and the first principal component separated the disease/healthy groups almost perfectly (red - disease, blue - healthy):

1 http://dijkstra.cs.ttu.ee/~Karl.Marka/PCA.png

I extracted the weights for the first principal component and ranked the probes according to their weights. It turns out that ALL the probes, that have a positive weight are downregulated in the disease population and the probes with negative weights are upregulated.

2 http://dijkstra.cs.ttu.ee/~Karl.Marka/weights.jpg

How does one interpret that result? Does it mean that the best predictors for disease are genes that are downregulated? Or does it merely mean that the PCA figured out that the best predictor is the up- or downregulatedness of a gene and that this selection of 135 probes together are able to separate the two groups?

$\endgroup$

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.