近年來藥物、人體受器蛋白質以及藥效的資訊快速累積，又因抗藥性和副作用使得單一藥物逐漸不敷使用，而新藥開發則需要面對龐大的實驗組合，因此電腦篩選(in silico prediction) 成為了藥物開發過程的一環，透過此方法可以事先篩選出較有可能產生反應的實驗組合，節省實驗時間及成本的消耗。我們提出一種方法來分析”藥物–目標蛋白質”的反應網路，藉由這樣的方法可以預測未實驗過的藥物-目標蛋白質組合是否會產生反應，以及找出藥物-目標蛋白質所擁有的底層官能基及化學結構之前的關聯性。為了方便與其他論文比較，實驗使用Yoshihiro Yamanishi(et al.,2010) 所提供的資料庫來進行實驗。最後在實驗結果中顯示，我們所提出的方法在經由適當的權重分配調整之後，可以非常接近比較的對象中最佳方法的準確度，藉由這個準確度的測試可以使我們所推論出的底層官能基及化學結構之前的關聯性有一定的可信度。

In recently years, because the quantity of drug and human protein information increase quickly, the nova drug development has to face to large possible drug-target experiment pair. Therefore, the in silico prediction method become important step in drug development to cut down the cost and have raised much attention. Through those prediction methods that we can filter out the possible drug-target pair before actually conducting biological experiments or even human test.
We proposed a new method based on network method and machine learning method to predict high probability interactional drug-target pair. Furthermore, we also want to find out the associations between the features of drug and features of target. We use the data integrated by Yoshihiro Yamanishi(et al.,2010) for comfortably to compare accuracy.
In the result of experiment, we construct a bipartite network of both the chemical structure and protein domain association networks. The accuracy performance of our method approaches to that of the best method very closely. This result can provide confidence that our c association network actually helps in revealing the association between the two heterogeneous data features.

1. Introduction
1.1 Background
1.2 Relative work
1.3 Motivation
2. Methods
2.1 Objective
2.2 Material
2.3 Optimization problem
2.4 Search of optimal H
3. Experiments and Results
3.1 Experiments setting
3.2 Experiment 1 Result
3.3 Experiment 2 Result
3.4 Experiment 3 Result
4. Conclusions and Discussion
5. Reference

[1] Yamanishi Y, Araki M, Gutteridge A, Honda W, Kanehisa M (2008)
Prediction of drug-target interaction networks from the integration of chemical
and genomic spaces. Bioinformatics 24: i232–240.
[2] Yoshihiro Yamanishi, Edouard Pauwels, Hiroto Saigo, Veronique Stoven (2011)
Extracting sets of chemical substructures and protein domains governing drug-target interactions . Chem. Inf .Model.
[3] H. W. Kuhn and A. W. Tucker (1951) Nonlinear Programming. Berkely:
University of California Press
[4] Zhisong He , Jian Zhang , Xiao-He Shi , Le-Le Hu , Xiangyin Kong , Yu-Dong Cai ,
Kuo-Chen Chou (2010) Predicting Drug-Target Interaction Networks Based on Functional Groups and Biological Features. PLoS one
[5] Adams CP,Brantner VV:Estimation (2006) the cost of new drug development: Is it really 802 million dollars? Health Aff.
[6] D. Kim, S. Sra, and I.S. Dhillon, (2008) Fast projection-based methods for the least squares nonnegative matrix approximation problem. Stat. Anal. Data Min., Vol. 1(1),pp. 38-51.
[7] Maggio, E. T. and Ramnarayan, K. (2001). Recent Developments in Computational
Proteomics. Drug Disc. Today 6, 996-1004.