Predict Gene Expression Using Logistic Regression

• Activity Number: 293
• Type: Contributed
• Date/Time: Tuesday, August 8, 2006 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Lei Guo*+ and Yuan Yuan and Lei Shen and Jun Liu
• Companies: Harvard University and Harvard University and GlaxoSmithKline and Harvard University
• Keywords: logistic regression ; gene regulation ; transcription factor binding sites ; cross validation

Abstract

Transcription factors (TFs) play crucial roles in gene regulation by interacting with genomic DNA. It has been shown that the Bayesian network can be applied to learn regulatory network structures and predict expression patterns of genes by their sequence information (Beer and Tavazoie, 2004). Although the prediction accuracy is 73%, it is not clear whether other simpler models can do better. We experimented with a simple logistic regression model together with variable selection, and achieved higher prediction accuracy (78%) under the setting identical to that of Beer and Tavazoie, even without motif site orientation and location information. Furthermore, we showed how their incorrect cross-validation inflated the estimated prediction accuracy.