source: plugins/base1/se.lu.thep.wenni/trunk/base/base1/base_plugin_script/plugin_WeNNI.base.in @ 826

Last change on this file since 826 was 826, checked in by Jari Häkkinen, 13 years ago

Fixes #148.

File size: 3.3 KB
Line 
1BASEfile
2section plugin
3uniqueName  thep.lu.se/jari/wenni
4versionNumber MAKESUBSTOFVERSIONNUMBER
5name  Transformation: WeNNI
6descr Weighted nearest neighbour imputation, WeNNI.\r\n\r\nWeNNI is a method for imputing unreliable expression data values. WeNNI imputes the missing values by using continuous spot quality weights. In WeNNI the quality of data values are not divided into groups of either missing or present, but rather a continuous quality weight is assigned to each data value.\r\n\r\nThe imputation performed by WeNNI is controlled with two parameters: the number of nearest neighbours to consider in the calculations and beta that is used in the calculation of weights. In the present version of WeNNI, the quality weights are calculated from a signal-to-noise ratio. The results presented in the WeNNI publication (referenced below) show that the choice of parameters is not crucial, and a value around 10 as the number of nearest neighbours and a beta in the range 0.1 to 1 is suggested when parameter tuning cannot be performed.\r\n\r\nWeNNI is described in "Improving missing value imputation of microarray data by using spot quality weights", P. Johansson and J. H\344kkinen, BMC Bioinformatics 7, 306 (2006). In this paper, we assessed imputation methods on three data sets containing replicate measurements. Of the compared methods, best performance, and robustness were achieved with WeNNI.\r\n\r\nComments on WeNNI:\r\n\r\n1) The notion of weights becomes obsolete after running WeNNI, i.e., do not use the weight fed into WeNNI in any subsequent analysis because all weights are now strictly 1.\r\n\r\n2) Running WeNNI as a BASE plug-in makes WeNNI destined to impute log ratios of channel 1 and channel 2 (M values in BASE world). A consequence of imputing log ratios is that a change in ratio cannot be assigned to a specific channel. This implies that log(channel1*channel2) (A values in BASE world) become undefined and useless. However, on request from BASE users it was decided that A values should not be affected by the transformation in cases where the A value is well defined before imputation. In cases when an A value do not exist before transformation (i.e. channel1<=0 or channel2<=0) it was decided that A should be set to 0. NOTE, this does not change the underlying WeNNI algorithm in any way but is rather conventions needed for BASE plug-in usage.\r\n\r\n3) It is recommended to filter reporter on presence before running WeNNI. The filter can be liberal, say 50%. You can perform the presence filter in the "Gene filter" box above. Select "In # of Assays".\r\n\r\nFeedback, suggestions for improvements, criticism, or questions regarding the algorithm should be sent to Peter Johansson, peter\@thep.lu.se, or Jari H\344kkinen, jari\@thep.lu.se.
7execName  MAKESUBSTOFEXECNAME
8geneAverages  0
9serialFormat  0
10url http://baseplugins.thep.lu.se/wiki/se.lu.thep.WeNNI
11minChannels 2
12maxChannels 2
13leaveStdin  0
14leaveStdout 0
15estimatedTime 3600
16defaultMaxRam 134217728
17usedColumns position\treporter
18usedFields  intensity1\tintensity2\tBCh1SD\tBCh2SD
19columns position  valueType name  commonName  options defaultValue  enumOptions removed
20%
211 h section   30  WeNNIParams   0
222 f beta  Beta  30  0.6   0
233 i neighbours  Neighbours  30  4   0
244 h intensity1    30  intensity1    0
255 h intensity2    30  intensity2    0
266 h BCh1SD    30  BCh1SD    0
277 h BCh2SD    30  BCh2SD    0
28
Note: See TracBrowser for help on using the repository browser.