02.08.2013 Views

Sample A: Cover Page of Thesis, Project, or Dissertation Proposal

Sample A: Cover Page of Thesis, Project, or Dissertation Proposal

Sample A: Cover Page of Thesis, Project, or Dissertation Proposal

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

probes can thus be reinc<strong>or</strong>p<strong>or</strong>ated into the analysis, as the investigat<strong>or</strong> deems appropriate. The<br />

impact <strong>of</strong> each variable is study dependent, but, since our interest is to identify diagnostic<br />

signatures, there is a requirement f<strong>or</strong> gene patterns that are robust to individual sampling and<br />

technical variation, allowing high accuracy in sample classification, whether binary <strong>or</strong> multistate<br />

[31-35].<br />

An advantage <strong>of</strong> the multi-probe per transcript platf<strong>or</strong>ms is that multiple measurements are<br />

available per gene-sample, increasing confidence in the measurement [36]. To retain this feature<br />

is imp<strong>or</strong>tant, so, after BaFL filters out probes subject to confounding fact<strong>or</strong>s, our analysis pipeline<br />

currently requires that a minimum <strong>of</strong> 4 probes, per ProbeSet, per array, must be present. An<br />

optional constraint is that this must be exactly the same 4 probes per array. Variation due to the<br />

technical complexity <strong>of</strong> the assay is completely lab dependent [6, 8, 20] and cannot be quantified<br />

in the same way as the fact<strong>or</strong>s listed above, so simple statistical tests are used. Two simple tests<br />

<strong>of</strong> overall similarity are: the total amount <strong>of</strong> response, assessed by measuring the total amount <strong>of</strong><br />

label present, and the total number <strong>of</strong> genes contributing to that response, assessed from the total<br />

number <strong>of</strong> probes giving signal. This can be determined at both the probe and ProbeSet level. In<br />

these tests, care must be taken only to use signals that can be directly compared, so the<br />

manufacturer’s specification f<strong>or</strong> the linear range <strong>of</strong> the scanner is used to set lower and higher<br />

bounds f<strong>or</strong> interpretable signal [16, 37]. It is possible that these criteria exclude samples<br />

unnecessarily, but in the absence <strong>of</strong> calibration standards and consistent controls we consider this<br />

to be a reasonable, conservative approach [18].<br />

The remaining probes can be collected either as a linked set <strong>of</strong> values <strong>or</strong> aggregated into a single<br />

transcript- <strong>or</strong> gene-level value (similar to ‘ProbeSet’ values computed by other alg<strong>or</strong>ithms).<br />

29

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!