Matched vs tumor_only scores

[bounlu]

I have compared MSI scores for the same tumor samples between:

1. its matched normal (msi)
2. a baseline of normals (pro)

I had 65 tumor and 58 normal samples of which 55 are matched.

In case 1 above, I used 55 matched samples and run the analysis for each normal-tumor pair.

In case 2, I used all 58 normal samples to create a baseline and compared each of tumor samples to this baseline.

I used default values for all parameters.

Between the two analysis, I got very different results for the same tumor samples in terms of somatic sites / MSI scores.

Case 1 gave none to very few events, resulting in around 0% MSI. Case 2 gave a range more than 1%, with a few samples up to 5-10% MSI.

How to explain this discrepancy? Which one to trust more?