QPROT / Blog: Recent posts

There was a bug in the data reading code in getfdr function. Some proteins may have infinity Z statistics (inf, -inf), and these numbers will be set to a large number (1000, -1000 respectively), and will be the most significant hit in our list. (courtesy of Dean Hammond)

QPROT v1.3.1 had a critical bug in the code that led to conservative estimation of FDR. The bug has been fixed and tested in several gold standard datasets.

There is still a possibility that the global FDRs (FDRup and FDRdown column) can be miscalculated due to numerical underflow -- in which case use localFDR (fdr column) for filtering.

Most functionalities have not changed.

Update #1.
If the proportion of differentially expressed proteins is too small, it will automatically correct it to 1%.

Update #2.
The model fitting progress (# of iterations) will be reported less frequently (every 1000th iteration, as opposed to every 100th iteration)

Update #3.
The mixture density of Z-statistics (null distribution, alternative distribution, and combined distribution) will be scaled to their proportion in the data. Previously the raw density was reported for each mixture component (second and third column) without relative weighting. The new version will report scaled components, so that the second and third column will add up to the fourth column when the data is plotted.