Control contamination

By reducing the cutoff, for example to p < 0.01, you can decrease the number of false positives even for the large chip. However, you can do better using more efficient and more complicated methods of multiple testing. 'Doing better' means use a tighter filter complying to the same standard. A classical but very rigid standard is the family-wise error rate (FWER). Essentially, it does not tolerate false positives in your list at all. While it cannot completely ensure a perfectly clean list, it can do so at least with a high probability. Hence if you set the FWER to 0.05, there is only a 5% chance that a single false-positive gene could sneak through the filter into the list. For a good introduction into the FWER see Dudoit et al. (12). In R, function mt.rawp2adjp in package multtest offers several procedures to build FWER filters. For the classical Bonferroni-Holm procedure, type:


FWER <- mt.rawp2adjp(pvalue,proc="Holm") Be less restrictive

FWER based standards are about the highest available. But it might happen that not a single gene passes the filter. While this is a 'spotless list', it is probably not what you have hoped for. If you are willing to tolerate some false positives in your list, say 5%, you can do better using the false discovery rate (FDR) introduced by Benjamini and Hochberg (13). In R type:

FDR <- mt.rawp2adjp(pvalue,proc="BH")

0 0

Post a comment