Good Day All,

I have new molecular-based diagnostic test which focuses on detecting SNPs in a 10 different genes. The DNA diagnostic test is actually probe based so it is looking at relatively small sequences (e.g. 16-24 base pairs long).

I would like to assess the test's overall accuracy as well as do an in-depth analysis on a set of well characterized clinical samples (perhaps a n~100-200) and am curious about thoughts on how the appropriate statistical test. The test is not a true sequencing 'platform' but it is rather similar.

In my opinion, there are several layers which I can assessed accuracy

  • At a macro level, if any nucleotide is wrong the test is wrong. Run 100 tests and calculate test sensitivity/specificity and use McNemar's Test (crudest).
  • At a micro level, look at each nucleotide (e.g. truth vs test) for the small sequences (therefore at a minimum 10*16=160 pairs per 'run') and then run the McNemar's on each run (so 160 pairs and 100 runs)
  • I am sure there are many other permutations so I am curious as to the appropriate statistical techniques to evaluate the accuracy, sensitivity and specificity of this test?

    Similar questions and discussions