Quantitatively integrating molecular structure and bioactivity profile evidence into drug-target relationship analysis
© Xu et al.; licensee BioMed Central Ltd. 2012
Received: 7 November 2011
Accepted: 23 March 2012
Published: 4 May 2012
Public resources of chemical compound are in a rapid growth both in quantity and the types of data-representation. To comprehensively understand the relationship between the intrinsic features of chemical compounds and protein targets is an essential task to evaluate potential protein-binding function for virtual drug screening. In previous studies, correlations were proposed between bioactivity profiles and target networks, especially when chemical structures were similar. With the lack of effective quantitative methods to uncover such correlation, it is demanding and necessary for us to integrate the information from multiple data sources to produce an comprehensive assessment of the similarity between small molecules, as well as quantitatively uncover the relationship between compounds and their targets by such integrated schema.
In this study a multi-view based clustering algorithm was introduced to quantitatively integrate compound similarity from both bioactivity profiles and structural fingerprints. Firstly, a hierarchy clustering was performed with the fused similarity on 37 compounds curated from PubChem. Compared to clustering in a single view, the overall common target number within fused classes has been improved by using the integrated similarity, which indicated that the present multi-view based clustering is more efficient by successfully identifying clusters with its members sharing more number of common targets. Analysis in certain classes reveals that mutual complement of the two views for compound description helps to discover missing similar compound when only single view was applied. Then, a large-scale drug virtual screen was performed on 1267 compounds curated from Connectivity Map (CMap) dataset based on the fused similarity, which obtained a better ranking result compared to that of single-view. These comprehensive tests indicated that by combining different data representations; an improved assessment of target-specific compound similarity can be achieved.
Our study presented an efficient, extendable and quantitative computational model for integration of different compound representations, and expected to provide new clues to improve the virtual drug screening from various pharmacological properties. Scripts, supplementary materials and data used in this study are publicly available at http://lifecenter.sgst.cn/fusion/.
To comprehend relationship between intrinsic characteristics of chemical compound and the compound interaction with protein target is an essential task to evaluate potential protein-binding function for virtual drug screening. Similarity relationship between compounds can be characterized differently, depending on different aspects of features to be measured. The similarity measurement of small molecules has been the focus of essentially every compound-based approach to design or identify novel drug candidates . However, in the process of novel drug screening, the representation of a compound varies dramatically, which results in different similarity measurements. Such similarity difference has given rise to distinct candidate compound similarity ranking lists with only generally about 15% overlap . It is demanding and necessary if information from multiple data sources can be integrated together to produce a comprehensive representation and assessment of similarity relationship between small molecules , thus expected to boost the results of virtual drug screening.
Generally, the drug candidates are related to specific targets. The investigation on the nature of target-specific structure–activity relationships of molecules should be based on the available data sources concerning structure, activity and target-binding information from a comprehensive and integrative perspective. Fortunately, public resources are in a rapid growth both in the quantity of data and in the type of data-generating, which provide us a great chance to further mine the relationship between compounds and their targets. Besides the classic representations of small molecules, like various fingerprints characterizing compound chemical structure, public high-throughput experimental data representing bioactivity of compounds are boosting with the development of online database, including PubChem (http://pubchem.ncbi.nlm.nih.gov/) , Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo/)  and DrugBank (DrugBank, http://drugbank.ca/)  etc., which provides an alternative way for molecule characterization based on bioactivity profiles. Several recent studies on the relationship between different compound features claimed that, correlations were proposed between bioactivity profiles and target networks, especially when chemical structures were similar [2, 6–8]. By simply combining both public repositories of compound targets and compound bioactivity, these studies indicates that comparison of bioactivity profile can provide insight into the mode of actions (MOA) at the molecular level, which will facilitate the knowledge-based discovery of novel compounds. However although various relationship were found between multiple features, no effective quantitative integrating methods was proposed or evaluated to combine these multi-view features. Inspired by previous works, two important and interesting computational issues are needed to investigate: (1) is there a quantitative relationship between compound features (bioactivity profile and structural feature) and compound target that can be specifically described? (2) Since the former works implicated that an integration of multiple compound features may result in a better measurement of target-specific compound similarity rather than only one specific type was adopted, how such integration can be optimized to quantitatively and automatically combine information from various views of compound representations, i.e., structural features, bioactivity features and other more? Hereby in our study, we refer such multiple features description and integration for compound as a multi-view data representation and learning problem, and we aim at presenting a quantitative relationship between target-specific compound similarity and multi-view representations of compound features in an efficient multi-view learning schema.
It should be noted that the term “multi-view learning” was initially presented from 3D-object recognition by the machine learning and graphic communities . Naturally as implicated by its name, multi-view learning combines models from different aspects of one identical entity to obtain an overall and comprehensive representation for further study. Multi-view learning was classically introduced as co-training, a semi-supervised learning procedure to distinguish webpages using two different types of data . Thereafter the concept of integration of different information sources has been developed for years in the field of information retrieval [11–13]. On the other side, as an unsupervised-learning method, multi-view clustering algorithms can be divided into two categories in general : (1) Fusion of similarity data by deriving a convex combination of similarities from different views to minimize a given penalty error [15, 16]. (2) Fusion of clustering decision derived from each view separately [17, 18]. In the clustering process, other techniques like canonical correlation analysis (CCA)  and matrix factorization  were employed to reduce the feature dimension or reconcile clustering groups. These applications of multi-view learning commonly yield better performance than that of single-view learning. In our study, as both the structure and bioactivity information are two distinguished intrinsic features to describe the small molecule, it is natural to investigate the results with the integration of both the chemical space (molecule structure) and genetic space (bioactivity profile) of molecules for a better evaluation of molecular properties and similarity comparison.
In this study, firstly a data set of 37 compounds (in Additional file 1: Table S1) from previous study based on bioactivity profile similarity  were adopted. Two similarity matrix characterizing bioactivity profile and structural similarity were calculated. As we would like to investigate the hierarchical structure of similarity among compounds regarding to multiple data sources, rather than only achieve an integrated ranking decision, a similarity fusion method was employed and modified to automatically optimize the weights of the combination of different similarity data. A hierarchy clustering was produced and discussed based on the fused similarity. Then, in order to evaluate the fusion method on the large scale dataset, Connectivity MAP dataset  containing 1267 compounds with their gene expression profile and structure fingerprint representation were used to perform drug virtual screen based on similarity searching. The compound-target interaction in these experiments was also analysed and compared quantitatively to demonstrate the benefits introduced by the integration of multiple data representations.
Materials and methods
In our study, the same data set used in Cheng’s work  rather than the up-to-date data is applied for equally comparison purpose, in order to illustrate the superiority of target-relationship analysis with similarity fusion from integration of multi-view information. The NCI-60 data set is available in the PubChem BioAssay Database, derived from the bioassays titled “NCI human tumor cell line growth inhibition assay” with relatively sufficient number of tested compounds (more than 16,000). Finally, filtered through 3 rules as Cheng defined , 37 small molecules of eligible quality were curated as the final NCI-60 dataset (in Additional file 1: Table S 1).
In order to demonstrate the performance of the feature integration on the large-scale dataset, similarity fusion was performed on the well-known Connectivity Map dataset.  Justin Lamb, et al. had created the first reference collection of gene-expression profiles from cultured human cells stimulated with bioactive small molecules, together with the pattern-matching algorithm to mine these data. To date, CMap contains approximately 7,100 expression profiles representing 1,309 compounds. Some compounds with only expression profiles of HT_HG-U133A_EA Gene chips were not included in this study due to the lack of chip description information. Compared to the former NCI-60 data, the gene expression profile for a compound can also be viewed as a kind of bioactivity representation.
Test for NCI-60 dataset
Similarity matrix from two views: Bioactivity profile and molecule structure
where is the number of features in compound A(B), and is the number of features common to both A and B. Both of the two similarity measurements are in the interval from 0 to 1. It should be noted that correlation coefficient of bioactivity profile below 0 are assign to 0 for two reasons: (1) only very few compounds pairs have a negative correlation coefficient and the minimum is −0.2, which is not significant as an evidence of negative correlation; (2) regarding to the integration analysis of different similarity information, negative correlation brings in no better information of molecular similarity than noise. Finally, as the input for multi-view fusion , the two similarity matrices were standardized as and renormalized to .
Fusion of similarity matrices by expectation-maximization (EM) algorithm
Non-negative matrix factorization:
Optimizations of weights for similarity matrices by minimizing the cross-entropy:
Hence the second step becomes a linear program problem.
Then the second step becomes a NLP problem and can be solved with LINDO API 6.1.
Average mean disagreement (AMD)
Average Dunn’s Index (ADI)
ADI is used to describe the partition quality in a clustering result. Dunn’s index is defined as the ratio of the minimal interclass distance and maximal intra-class distance. Higher Dunn’s index indicates better validity of partition. For NCI-60 dataset we calculate the average Dunn’s index regarding to a range of class number as defined in AMD. The η that obtains a high Dunn’s index with low variance could be considered as a proper estimation.
Fused similarity matrix
After estimating the sparseness controlling parameter, the alternative minimization steps were repeated on the whole dataset. Final weights vector α was calculated and the fused similarity matrix can be obtained as the convex combination of the two original similarity matrices with the weights α.
Compound-target interaction analysis
A compound-target interaction network can be constructed via target annotation in the PubChem BioAssay database. In general, one compound (identified by CID) was linked to a target protein (identified by NCBI protein ID, or GI) if this compound was tested active in the bioassay which was specified with the protein target. All the target annotation and activity information was retrieved from PubChem BioAssay database via E-Utilities tool. The interaction network was constructed and visualized by using the Cytoscape (version 2.7.0) , containing 37 compound nodes and 138 target nodes (in Additional file 1: Table S 2).
Similarity fusion on large-scale CMAP dataset
In order to demonstrate the performance of the feature integration on the large-scale dataset, similarity fusion was performed on the CMap dataset. In this study, in order to generate pair-wise relationship among all the compounds, Gene Ontology (GO) fingerprint , which is presented in our previous study as a well-defined bioactivity representation, was adopted to combine all the expression profiles of one compound and reduce the high dimensions and noises in the microarray data. This descriptor was used to describe drug in a biological activity view. Similarly, the same structural fingerprint as used for NCI-60 data was used here to describe drug in a compound structure view.
The fusion of structural fingerprint and GO-fingerprint similarity matrices was performed following the same workflow aforementioned for NCI-60 data. And the detailed parameter optimization will not be discussed here. Considering that clustering result of large scale dataset cannot be analysed straightforwardly as the former 37-compound dataset, two typical HDAC and HSP90 inhibitors, which was used as the examples in Lamb’s work, were chosen as the queries to validate our fusion method from the perspective of virtual drug screen. For each query, the ranks of similarity searching derived by the fused similarity were compared to that with only single view, and the targets of top-ranked compounds with similarity above 0.5 to queries were also analysed for further discussion.
Results and discussions
Test results for NCI-60 dataset
Assessment of the sparseness-controlling parameter for NCI-60 data
As shown in Figure 2A, the Average Mean Disagreement reached the lowest value when . Furthermore, the Average Dunn’s Index indicated the validity of the clustering. As shown in Figure 2B, the ADI grew gradually when η increased below 3. The decreasing variance suggested an accretive clustering quality. It should be noted that when ADI has a sharp rise, while after that the trend of growing has become attenuated. Later calculation of weights α reveals that η lower than 3 or greater than 100 will tend to give biased weights to the two matrix, i.e. either or . In summary, given the best value of η in AMD, and a relative high value in ADI, it is reasonable to choose as a proper estimation to control the sparseness.
Overall average common target number
Highly similar structure as complement of bioactivity profiles
Further compound-target interaction analysis on the protein targets within this cluster shows that the compound CID: 3246719 shares common protein target with other members in the group, while this compound was missing in the previous compound-target network  as shown in Additional file 1: Figure S 1A. It can be seen that there’re three protein target linked to compound CID: 3246719 [GI: 119579178, 222080095 and 3287985] (Figure 5). All the three protein targets are shared in the former single-view clustering. It is indicated that by using multi-view similarity analysis a missing group member was discovered by introducing extra structural information. It is evident that compounds sharing similar structural features and bioactivity profiles simultaneously will give bonus to the performance in a similarity-based search. In addition, other two compounds (Cluster C, [CID: 4212 and 5458171]) can be another good examples. These two compounds are significantly similar both in structure and bioactivity profiles (Additional file 1: Figure S 3); hence they get a notably high similarity in the hierarchy tree (Figure 1).
Highly similar bioactivity profiles as complement of moderately similar structure
Another cluster composed of three compounds [CID: 2723601, 3246652 and 5351879] are noteworthy to explain in the fused hierarchy clustering tree. If we only measure the compound similarity with structural information, we can see that there are relatively less similar. However, when combined with the bioactivity information, these three compounds successfully merged into the first cluster during hierarchical clustering . (Additional file 1: Figure S 2). Target interacting analysis reveals that these three compounds share a common target [GI: 4504349], indicating a potential common function in biological process. It is notable that certain fragment of the compounds, thioguanine in this example, instead of the complete structure, is essential in a binding event. Therefore when compounds that bind to a common target exhibit only relatively low overall structural similarity, it could be a good complementary to introduce the bioactivity profiles to suggest a more strong correlation with target binding potent. Such advantage of multi-view similarity assessment could be remarkable when no prior knowledge about either specific functional fragment or target is available.
Drug virtual screen based on fused similarity of CMap dataset
Similarity search based on fused similarity: HDAC inhibitors
Similarity search based on fused similarity: HSP90 inhibitors
15-delta prostaglandin J2
15-delta prostaglandin J2
A multi-view clustering method was introduced to discover a more robust correlation between fused multi-view similarity and compound-target interacting pattern. By using a similarity-based optimization and fusion model, a hierarchy clustering integrated with both structural and bioactivity profile information was presented on the NCI-60 dataset. It is interesting that comparing to single view analysis, the overall common target number within fused classes has been promoted by integrating information from two views, which indicated a more robust and efficient representation of compound related to specific target. Analysis of compound-target interaction network shows that fusion of data source from different views enhances similar compound discovery, leading to a more comprehensible assessment of target-binding potent. Further analysis in certain classes with high fused similarity shows that the mutual complement of the two views can lead to the discovery of missing similar compound with only one view. A further large-scale similarity searching on the CMap data based on the fused similarity also obtained a better ranking results compared to that of single-view for two inhibitors as queries, thus indicate the potential use of our quantitative similarity fusion in virtual drug screen. In summary, our findings are interesting for the following reasons: Firstly, both the bioactivity profiles and structural fingerprint lack to be a completely direct indicator of interaction, i.e. only partial features instead of overall characterization from either view are essential in a binding event. Hence by integrating potentially correlating features from both views to maximize the utility of available data source, a robust similarity assessment could be achieved without prior knowledge about the detail relationship between target-binding rules and compound features. Secondly, the fusion method in this study provides an extendable framework of integrating multi-view data. Fusion process is applicable to various situations when more than two data sources are available. A comprehensive assessment of the similarity can be achieved in virtual drug screening when various potential pharmacological properties of compounds are integrated.
Funding: This work was supported in part by Project White Magnolia Funding, Shanghai (Grant No. 2010B127), Shanghai Pujiang Talents Funding (Grant No.11PJ1407400), Tongji Excellent Young Scientist Funding (Grant No. 2000219052), National Natural Science Foundation of China (Grant No. 30976611, Grant No.31100956 and Grant No. 61173117), Research Fund for the Doctoral Program of Higher Education of China (Grant No.20100072120050, 20110072120048) and 973 National Key Basic Research Program of China (Grant No. 2010CB833601).
We would like to express our special thanks for Tiejun Cheng and other authors of  in NIH, U.S. They kindly helped us for the NCI-60 dataset retrieving.
- Eckert H, Bojorath J: Molecular similarity analysis in virtual screening: foundations, limitations and novel approaches. Drug Discov Today 2007, 12(5–6):225–233.View ArticlePubMed
- Cheng T, Li Q, Wang Y, Bryant SH: Identifying compound-target associations by combining bioactivity profile similarity search and public databases mining. Journal of Chemical Information and Modeling 2011, 51(9):2440–2448. 10.1021/ci200192vPubMed CentralView ArticlePubMed
- Bryant SH, Wang YL, Xiao JW, Suzek TO, Zhang J, Wang JY: PubChem: a public information system for analyzing bioactivities of small molecules. Nucleic Acids Res 2009, 37: W623-W633. 10.1093/nar/gkp456PubMed CentralView ArticlePubMed
- Barrett T, Troup DB, Wilhite SE, Ledoux P, Evangelista C, Kim IF, Tomashevsky M, Marshall KA, Phillippy KH, Sherman PM, et al.: NCBI GEO: archive for functional genomics data sets-10 years on. Nucleic Acids Res 2011, 39: D1005-D1010. 10.1093/nar/gkq1184PubMed CentralView ArticlePubMed
- Wishart DS, Knox C, Guo AC, Cheng D, Shrivastava S, Tzur D, Gautam B, Hassanali M: DrugBank: a knowledgebase for drugs, drug actions and drug targets. Nucleic Acids Res 2008, 36: D901-D906.PubMed CentralView ArticlePubMed
- Cheng TJ, Wang YL, Bryant SH: Investigating the correlations among the chemical structures, bioactivity profiles and molecular targets of small molecules. Bioinformatics 2010, 26(22):2881–2888. 10.1093/bioinformatics/btq550PubMed CentralView ArticlePubMed
- Sheng Z, Kang H, Dai TL, Liu Q, Zhu RX: Complementary study of structure features and gene profile features for chemical compounds. Acta Chim Sinica 2010, 68(23):2395–2400.
- Glover CJ, Rabow AA, Isgor YG, Shoemaker RH, Covell DG: Data mining of NCI's anticancer screening database reveals mitochondrial complex I inhibitors cytotoxic to leukemia cell lines. Biochem Pharmacol 2007, 73(3):331–340. 10.1016/j.bcp.2006.10.005PubMed CentralView ArticlePubMed
- Hartley R, Zisserman A: Multiple view geometry in computer vision. Cambridge University Press, The Edinburgh Building, Cambridge, CB2 2RU, UK; 2003.
- Blum A, Mitchell T: Combining labeled and unlabeled data with co-training. In Proceedings of the eleventh annual conference on Computational learning theory. Madison, Wisconsin, United States: ACM; 1998: 92–100 1998: 92-100
- Ghani R: Combining labeled and unlabeled data for multiclass text categorization. In Proceedings of the Nineteenth International Conference on Machine Learning. Morgan Kaufmann Publishers Inc, San Francisco, CA, USA; 2002: 187–194 2002: 187-194
- Brefeld U, Scheffer T: Co-EM support vector learning. In: Proceedings of the twenty-first international conference on Machine learning Banff, Alberta, Canada: ACM; 2004:16. Banff, Alberta, Canada: ACM; 2004:16.
- Zhou Z, Li M: Semi-supervised regression with co-training. In: In Proc of Int’l Joint Conf on Artificial Intelligence, Sydney, Australia 2005.
- Bruno E, Marchand-Maillet S: Multiview Clustering: a late fusion approach using latent models. Proceedings 32nd Annual International Acm Sigir Conference on Research and Development in Information Retrieval 2009. Boston, MA, USA: 736–737.870. Boston, MA, USA: 736-737.870.
- Lange T, Buhman J: Fusion of similarity data in clustering. In Proceeding of Advances in Neural Information Processing Systems 18. MIT Press, Cambridge, MA, USA; 2005.
- Long B, Yu PS, Zhang ZM: A general model for multiple view unsupervised learning. In In SDM. Atlanta, Georgia, USA; 2008.
- Cowan I, Tesauro G, Virginia R, De Sa: Learning classification with unlabeled data. Proc NIPS93 Neural Information Processing Systems, Morgan Kaufmann Publishers, Denver, CO, USA, 1993, Pages: 112–119.
- Bickel S, Scheffer T: Multi-view clustering. In Proceeding of IEEE data mining Conference. Brighton, UK, 2004; 2004.
- Chaudhuri K, Kakade SM, Livescu K, Sridharan K: Multi-view clustering via canonical correlation analysis. In Proceedings of the 26th Annual International Conference on Machine Learning. Montreal, Quebec, Canada: ACM; 2009:129–136.
- Greene D, Cunningham P: A matrix factorization approach for integrating multiple data views. In Proceedings of the European Conference on Machine Learning and Knowledge Discovery in Databases: Part I. Springer, Bled, Slovenia; 2009:423–438 2009:423-438
- Lamb J, Crawford ED, Peck D, Modell JW, Blat IC, Wrobel MJ, Lerner J, Brunet JP, Subramanian A, Ross KN, et al.: The Connectivity Map: using gene-expression signatures to connect small molecules, genes, and disease. Science 2006, 313(5795):1929–1935. 10.1126/science.1132939View ArticlePubMed
- Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, Amin N, Schwikowski B, Ideker T: Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res 2003, 13(11):2498–2504. 10.1101/gr.1239303PubMed CentralView ArticlePubMed
- Batagelj V, Mrvar A: Pajek - Analysis and visualization of large networks. In: Graph Drawing Software, Mathematics and Visualization.Edited by Jünger, Michael and Mutzel, Petra.Springer Berlin Heidelberg 2004:77–103 2004:77-103
- Csardi G, Nepusz T: The igraph software package for complex network research. InterJournal 2006, 1695. Volume: Complex Systems Volume: Complex Systems
- Team RDC: R: A language and environment for statistical computing. Austria: R Foundation for Statistical Computing, Vienna; 2009.
- Sheng Z, Huang Q, Kang H, Liu Q, Cao ZW, Zhu RX: A new fingerprint of chemical compounds and its application to drugs virtual screening. Acta Chim Sinica 2011, 69(16):1845–1850.
- Vanhaecke T, Papeleu P, Elaut G, Rogiers V: Trichostatin A - like hydroxamate histone deacetylase inhibitors as therapeutic agents: toxicological point of view. Curr Med Chem 2004, 11(12):1629–1643. 10.2174/0929867043365099View ArticlePubMed
- Ree AH, Dueland S, Folkvord S, Hole KH, Seierstad T, Johansen M, Abrahamsen TW, Flatmark K: Vorinostat, a histone deacetylase inhibitor, combined with pelvic palliative radiotherapy for gastrointestinal carcinoma: the Pelvic Radiation and Vorinostat (PRAVO) phase 1 study. Lancet Oncol 2010, 11(5):459–464. 10.1016/S1470-2045(10)70058-9View ArticlePubMed
- Keen JC, Yan L, Mack KM, Pettit C, Smith D, Sharma D, Davidson NE: A novel histone deacetylase inhibitor, Scriptaid, enhances expression of functional estrogen receptor alpha (ER) in ER negative human breast cancer cells in combination with 5-aza 2 '-deoxycytidine. Breast Cancer Res Tr 2003, 81(3):177–186. 10.1023/A:1026146524737View Article
- Li YH, Tao PZ, Liu YZ, Jiang JD: Geldanamycin, a ligand of heat shock protein 90, inhibits the replication of herpes simplex virus type 1 in vitro. Antimicrob Agents Ch 2004, 48(3):867–872. 10.1128/AAC.48.3.867-872.2004View Article
- Yao JQ, Liu QH, Chen X, Yang Q, Xu ZY, Hu F, Wang L, Li JM: Hsp90 inhibitor 17-allylamino-17-demethoxygeldanamycin inhibits the proliferation of ARPE-19 cells. J Biomed Sci 2010, 17:30.
- Ramanathan RK, Egorin MJ, Erlichman C, Remick SC, Ramalingam SS, Naret C, Holleran JL, TenEyck CJ, Ivy SP, Belani CP: Phase I Pharmacokinetic and Pharmacodynamic Study of 17-dimethylaminoethylamino-17-demethoxygeldanamycin, an inhibitor of heat-shock protein 90, in patients with advanced solid tumors. J Clin Oncol 2010, 28(9):1520–1526. 10.1200/JCO.2009.25.0415PubMed CentralView ArticlePubMed
- Wang SH, Xu YQ, Maine EA, Wijeratne EMK, Espinosa-Artiles P, Gunatilaka AAL, Molnar I: Functional characterization of the biosynthesis of Radicicol, an Hsp90 inhibitor resorcylic acid lactone from chaetomium chiversii. Chem Biol 2008, 15(12):1328–1338. 10.1016/j.chembiol.2008.10.006View ArticlePubMed
- Olmos G, Arenas M, Bienes R, Calzada M, Aragonés J, Garcia-Bermejo M, Landazuri M, Lucio-Cazaña J: 15-Deoxy-Δ<sup>12,14</sup>-prostaglandin-J<sub>2</sub>reveals a new pVHL-independent, lysosomal-dependent mechanism of HIF-1α degradation. Cellular and Molecular Life Sciences 2009, 66(13):2167–2180. 10.1007/s00018-009-0039-xView ArticlePubMed
- Stamatakis K, Sanchez-Gomez FJ, Perez-Sala D: Identification of novel protein targets for modification by 15-deoxy-Delta12,14-prostaglandin J2 in mesangial cells reveals multiple interactions with the cytoskeleton. J Am Soc Nephrol 2006, 17(1):89–98.View ArticlePubMed
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.