437 – Statistical Issues Specific to Therapeutic Areas III
An Ensemble of Classifiers for Time Course Classification of Response to Treatment in Psoriatic Patients
Joel Correa da Rosa
Rockefeller University
Sandra Garcet
Rockefeller University
Jaehwan Kim
The Rockefeller University, Laboratory of Investigative Dermatology,
Suyan Tian
The Rockefeller University, Laboratory of Investigative Dermatology
James G. Krueger
The Rockefeller University, Laboratory of Investigative Dermatology
Mayte Suarez Farinas
Rockefeller University
Psoriasis is caused by the complex interaction of genetic, environmental and immunological factors. In the last 10 years there has been an explosion of drug testing in Psoriasis with more than 35 new drugs in different phases of drug development. In this study the main goal is to build a genomic classifier to select, based on earlier time points genes and pathways that can predict in psoriatic patients response-to-treatment at week 12 or later. To develop the classifier, we have transformed longitudinal skin expression profiles from patients treated with 7 different drugs into time course scores using LDA-projections. By using Gene Set Variation Analysis (GSVA) we evaluated activity for some known psoriasis pathways. A ensemble of classifiers was built based on 5 methods: PLS, LDA, GLMnet, TGDR and PAM. Results from 500 bootstrapped samples pointed out that accuracy is increasingly gained as time points are incorporated in the time-course score. Classification accuracy at baseline was not better than random for most of the cases. More than 90% of accuracy in predicting response to treatment is achieved at 1st week for the specific-treatment classifier.