TY - JOUR
T1 - Efficient Drug-Pathway Association Analysis via Integrative Penalized Matrix Decomposition
AU - Li, Cong
AU - Yang, Can
AU - Hather, Greg
AU - Liu, Ray
AU - Zhao, Hongyu
N1 - Funding Information:
This study was supported by NIH grants GM59507 and CA154295, US NSF grant DMS 1106738, and Hong Kong Baptist University grant FRG2/14-15/069, RGC Ref No. 22302815 from the Hong Kong Research Grant Council. The authors thank the Yale University High Performance Computing Center (funded by NIH RR19895) for the computation resource and data storage.
PY - 2016/5/1
Y1 - 2016/5/1
N2 - Traditional drug discovery practice usually follows the 'one drug - one target' approach, seeking to identify drug molecules that act on individual targets, which ignores the systemic nature of human diseases. Pathway-based drug discovery recently emerged as an appealing approach to overcome this limitation. An important first step of such pathway-based drug discovery is to identify associations between drug molecules and biological pathways. This task has been made feasible by the accumulating data from high-throughput transcription and drug sensitivity profiling. In this paper, we developed 'iPaD', an integrative Penalized Matrix Decomposition method to identify drug-pathway associations through jointly modeling of such high-throughput transcription and drug sensitivity data. A scalable bi-convex optimization algorithm was implemented and gave iPaD tremendous advantage in computational efficiency over current state-of-the-art method, which allows it to handle the ever-growing large-scale data sets that current method cannot afford to. On two widely used real data sets, iPaD also significantly outperformed the current method in terms of the number of validated drug-pathway associations that were identified.
AB - Traditional drug discovery practice usually follows the 'one drug - one target' approach, seeking to identify drug molecules that act on individual targets, which ignores the systemic nature of human diseases. Pathway-based drug discovery recently emerged as an appealing approach to overcome this limitation. An important first step of such pathway-based drug discovery is to identify associations between drug molecules and biological pathways. This task has been made feasible by the accumulating data from high-throughput transcription and drug sensitivity profiling. In this paper, we developed 'iPaD', an integrative Penalized Matrix Decomposition method to identify drug-pathway associations through jointly modeling of such high-throughput transcription and drug sensitivity data. A scalable bi-convex optimization algorithm was implemented and gave iPaD tremendous advantage in computational efficiency over current state-of-the-art method, which allows it to handle the ever-growing large-scale data sets that current method cannot afford to. On two widely used real data sets, iPaD also significantly outperformed the current method in terms of the number of validated drug-pathway associations that were identified.
KW - drug discovery
KW - high dimensional data
KW - latent factor model
KW - matrix decomposition
KW - penalized method
KW - template
UR - http://www.scopus.com/inward/record.url?scp=84976529906&partnerID=8YFLogxK
U2 - 10.1109/TCBB.2015.2462344
DO - 10.1109/TCBB.2015.2462344
M3 - Journal article
C2 - 27295636
AN - SCOPUS:84976529906
SN - 1545-5963
VL - 13
SP - 531
EP - 540
JO - IEEE/ACM Transactions on Computational Biology and Bioinformatics
JF - IEEE/ACM Transactions on Computational Biology and Bioinformatics
IS - 3
M1 - 7172456
ER -