TY - JOUR
T1 - On predicting epithelial mesenchymal transition by integrating RNA-binding proteins and correlation data via L 1/2 -regularization method
AU - Qiu, Yushan
AU - Jiang, Hao
AU - Ching, Wai Ki
AU - Ng, Michael K.
N1 - Funding Information:
The authors would like to thank the four anonymous referees for their helpful and constructive suggestions. This research is supported in part of Natural Science Foundation of SZU (Grant No. 2017058), National Natural Science Foundation of China NSFC (Grant No. 91730301), Research Grants Council of Hong Kong under grant number 15210815, IMR and RAE Research Fund from Faculty of Science, the University of Hong Kong.
Funding Information:
The authors would like to thank the four anonymous referees for their helpful and constructive suggestions. This research is supported in part of Natural Science Foundation of SZU (Grant No. 2017058 ), National Natural Science Foundation of China NSFC (Grant No. 91730301 ), Research Grants Council of Hong Kong under grant number 15210815 , IMR and RAE Research Fund from Faculty of Science, the University of Hong Kong .
PY - 2019/4
Y1 - 2019/4
N2 - Identifying tumor metastasis signatures from gene expression data at the whole genome level remains an arduous challenge, particularly so when the number of genes is huge and the number of experimental samples is small. We focus on the prediction of the epithelial-mesenchymal transition (EMT), which is an underlying mechanism of tumor metastasis, here, rather than on tumor metastasis itself, to avoid confounding effects of uncertainties derived from various factors. We apply an extended LASSO model, L 1/2 -regularization model, as a feature selector, to identify significant RNA-binding proteins (RBPs) that contribute to regulating the EMT. We find that the L 1/2 -regularization model significantly outperforms LASSO in the EMT regulation problem. Furthermore, remarkable improvement in L 1/2 -regularization model classification performance can be achieved by incorporating extra information, specifically correlation values. We demonstrate that the L 1/2 -regularization model is applicable for identifying significant RBPs in biological research. Identified RBPs will facilitate study of the underlying mechanisms of the EMT.
AB - Identifying tumor metastasis signatures from gene expression data at the whole genome level remains an arduous challenge, particularly so when the number of genes is huge and the number of experimental samples is small. We focus on the prediction of the epithelial-mesenchymal transition (EMT), which is an underlying mechanism of tumor metastasis, here, rather than on tumor metastasis itself, to avoid confounding effects of uncertainties derived from various factors. We apply an extended LASSO model, L 1/2 -regularization model, as a feature selector, to identify significant RNA-binding proteins (RBPs) that contribute to regulating the EMT. We find that the L 1/2 -regularization model significantly outperforms LASSO in the EMT regulation problem. Furthermore, remarkable improvement in L 1/2 -regularization model classification performance can be achieved by incorporating extra information, specifically correlation values. We demonstrate that the L 1/2 -regularization model is applicable for identifying significant RBPs in biological research. Identified RBPs will facilitate study of the underlying mechanisms of the EMT.
KW - Classification
KW - Epithelial-mesenchymal transition (EMT)
KW - L -regularization
KW - RNA-binding proteins (RBPs)
UR - http://www.scopus.com/inward/record.url?scp=85055125740&partnerID=8YFLogxK
U2 - 10.1016/j.artmed.2018.09.005
DO - 10.1016/j.artmed.2018.09.005
M3 - Journal article
C2 - 30352711
AN - SCOPUS:85055125740
SN - 0933-3657
VL - 95
SP - 96
EP - 103
JO - Artificial Intelligence in Medicine
JF - Artificial Intelligence in Medicine
ER -