Introduction to OLIN package

Microarray measurements are affected by a variety of systematic experimental errors limiting the accuracy of data produced. Two well-known systematic errors for two-colour arrays are the so-called intensity-dependent and spatial (location-dependent) dye bias. Normalisation aims to correct for systematic errors in microarray data. The common linear (or global) normalisation method often fails to correct for dye bias as this bias is usually not linear. Although non-linear normalisation procedures have been able to reduce the systematic errors, these methods are based on default parameters and leave it to the user to choose “good” parameters. The optimal adjustment of the normalisation models to the data, however, can be crucial for the efficiency of the normalisation process [1]. The OLIN (Optimised Local Intensity-dependent Normalisation) R-package includes two normalisation schemes based on iterative local regression and model selection. Both schemes aim to correct for intensityand location-dependent dye bias in microarray data. For model selection (parameter optimisation), generalized cross-validation (GCV) is used. Additionally, several procedures to assess the efficiencies of normalisation are implemented in the package. A graphical user interface for OLIN has been implemented in the package OLINgui, that can be downloaded from the OLIN webpage (http://itb.biologie.hu-berlin.de/∼futschik/software/R/OLIN/index.html).