ORIGINAL ARTICLES  MICROARRAY MEETING 2003: III CONVEGNO ITALIANO SULLA TECNOLOGIA
Segrate (MI), 9-10 Giugno 2003 

Minerva Biotecnologica 2003 December;15(4):217-22

Copyright © 2004 EDIZIONI MINERVA MEDICA

lingua: Inglese

Control of selection bias in microarray data analysis

Furlanello C., Serafini M., Merler S., Jurman G.

ITC-IRST, Trento, Italy

We ­present an experi­men­tal set-up for the anal­y­sis and pre­dic­tion on micro­ar­ray ­data spe­cif­i­cal­ly ­designed to iden­ti­fy and cor­rect the ­impact of selec­tion ­bias in ­high-through­put prob­lems. A num­ber of recent­ly pub­lished and over­op­ti­mis­tic stud­ies ­present fea­ture selec­tion and ­gene pro­fil­ing pro­cess­es ­that ­incur over­fit­ting ­effects. We out­line the selec­tion ­bias prob­lem and dem­on­strate its ­effect on syn­thet­ic and micro­ar­ray ­data. We ­then intro­duce and ­describe a pro­ce­dure ­that suc­cess­ful­ly ­deals ­with the prob­lem ­through exten­sive resam­pling and ­label ran­dom­iza­tion tech­niques ­that ­employ sup­port vec­tor ­machines as a ­base clas­si­fi­er and an ­improved ver­sion of the recur­sive fea­ture elim­i­na­tion algo­rithm for ­gene rank­ing.