pl see the link

Genome-wide co-expression based prediction of differential ...

https://academic.oup.com/.../article/.../Genome-wide-co-expression-based-prediction-ofBased on the exploration of microarray data, many statistical methods have been .... expressionmeasurements for 6034 genes after data pre-processing (Singh et al., 2002). ... Based on ourexperiences, this method can generally improve the ..... of cancer: class discovery and class prediction by gene expression monitoring.

Hi I use combat to delete batch effect, but here you have different kind of arrays so first step would be a probeset by probeset mapping. I do a heatmap before and after batch effect removal to see whether the classes are more similar than the batches. Log2 is fine.

We in our group use often rank based methods for classification as well. The results for classification is medicore.

ComBat : https://www.bu.edu/jlab/wp-assets/ComBat/Abstract.html

Metanalysis tips: http://dx.doi.org/10.1093/nar/gkr1265

Thanks for the tips @ Dezső Módos!

Hi Hossein! Combat is the ideal function to control common "Non-biological experimental variations" to some extent. If you intend to combine one batch of files from diseased cohort and another one batch of files from non-diseased cohort, I would say "NO" for your classifier prediction analysis...There is no such a standard way to adjust the biological variations and proceed further analysis on the adjusted data.