Zhengdeng Lei, PhD

Zhengdeng Lei, PhD

2009 - Present Research Fellow at Duke-NUS, Singapore
2007 - 2009 High Throughput Computational Analyst, Memorial Sloan-Kettering Cancer Center, New York
2003 - 2007 PhD, Bioinformatics, University of Illinois at Chicago

Monday, June 4, 2012

two classifiers for cancer

In practice, we need two classifiers, one from microdissected tumor samples, the other from macrodissected tumor samples. Hence we can predicted microdissected samples by microdissected classifier, and macrodissected samples by macrodissected classifier.


Q: Is it not good to predicted microdissected samples by macrodissected classifier?
Answer: NOT good, confirmed.


Q: Really? It could also be due to population bias between training set and test set?
Answer: Not because of population bias


Q: it may be okay to predict macrodissected samples by microdissected classifier.?
Answer: NO, the other way around isn't work either.


 check with breast cancer datasets.

No comments:

Post a Comment