Printer Friendly

GEARS: A Genetic Algorithm Based Machine Learning Technique to Develop Prediction Models.

Byline: Wajahat M. Qazi, Zeeshan Iqbal, M. Saleem Khan, Muhammad Rehan, Jawaria Munir, Abdul Rauf Shakoori and Nasir-ud-Din

Abstract.- The development of new prediction models to identify potential modified residues are based on different machine learning methods. Primary sequences, biChemical properties of the amino acids and 3D structural information of proteins are used to evolve prediction models. The information about the significant residues to govern different biological prCesses has not been considered yet to develop a prediction model. MAPRes is an efficient tool which has been utilized to mine significant residues and assCiation patterns for surrounding amino acids of some specific modifications on hydroxyl and amino group such as phosphorylation and acetylation. The primary sequences of the proteins and assCiation patterns of surrounding amino acids of modified residues may use to train new dataset for the development of an efficient and reliable prediction model. Biophysical and biChemical properties of the amino acids are also important parameters for the prediction of the modified residues. This study proposes, GEARS (Genetic Evolution of ClAssifers by Learning Residue Rules and Sequences), a classifier rule learning model, which considered different machine learning techniques such as ANNs, HMM and MAPRes were considered for the development of GEARS model. The GEARS, by combining these models, will have the capacity to reduce the false negative and positive predictions.

Key words: Machine learning, bioinformatics, post translational modification, artificial neural networks, backpropagation, genetic algorithm, assCiation rules learning and mining, hidden markov models.

INTRODUCTION

Proteins are multi-functional molecules which often perform diverse functions (Jeffery,1999). The 3D structural changes are one of the significant factor behind the functional switch of proteins (Bork et al., 1998; Attwood, 2000; Li et al.,2004). These changes are regulated by a number ofbiChemical prCesses, particularly posttranslational modifications (PTMs) such as phosphorylation, sulfation, acetylation, methylation, glycosylation etc. (Varki and Kornfeld, 1980; Boynton et al.,2001; Berlot et al., 2002; Yuan et al., 2003). These PTMs are usually take place at amino group, hydroxyl group, carboxyl group and side chain of the amino acids. The amino group of the aminoacids modified by acetyl, methyl, formyl groups and etc. while modifications on hydroxyl group are O- Linked glycosylation and phosphorylation (Ahmad et al., 2006; Butt et al., 2011; Schiza et al., 2013). These modifications are vitally important to regulate various biological prCesses and the development of the prediction models to identify the potential modification sites for different types of modifications are equally important. To analyze the environment and potential modified sites for different PTMs several machine learning techniques and statistical models have been developed that are playing vital role to understand the structure- function relationship of proteins (Christlet and Veluraja, 2001; Blom et al., 2004; Fujii et al., 2004Lee et al., 2006; Qazi et al., 2006; Wong et al.,2007). The general classifications of these statisticaland computational models are prediction, classification and clustering techniques. Indeed these models have assisted biologists to provide information about the complex biological prCesses that are extremely difficult to find out utilizing existing experimental and/or theoretical studies. MAPRes (Mining AssCiation Patterns among preferred amino acid Residues in the vicinity of amino acids targeted for postTranslational modifications) is a useful computational model that can be utilize to mine assCiation rules/patterns for significantly preferred amino acids in the vicinity of specific PTM sites (Ahmad et al., 2008a). MAPRes is one of the example of clustering type protein sequence analyses.MAPRes has been utilized to mine assCiation rules/patterns for the surrounding amino acids that are biologically important in the vicinity of phosphorylation, glycosylation and acetylation sites (Ahmad et al., 2008a,b, 2009; Khawaja et al.,2008; Iqbal et al., 2013). The upgraded version of MAPRes has also been released (unpublished). This version of MAPRes provides mechanism to mine assCiation rules by utilizing the properties of neighboring amino acids. This version of MAPRes has been applied to study the acetylation sites based on the properties of surrounding amino acids such as polarity and charge (Iqbal et al., 2013). These studies have encouraged that assCiation rules can be used to build a better classifier by combining it with sequence learning.Indeed existing predictive and descriptiveanalyses methods are either based on data learning method (Blom et al., 2004) or data mining for assCiation analyses (Ahmad et al., 2008a). This suggests, in either case, these methods lack some significant aspects assCiated with data. Hence new methods are required to develop prediction models, utilizing the data derived from previous analyses in form of certain rules/patterns coupled with protein sequence data, by developing an assembly of learning schemes for different types of data. The concept of combinatorial classification for the development of prediction models has already been practiced such as DictyOGlyc (Gupta et al., 1999). In this prediction method, the jury of 6 ANNs with different configurations and architectures were utilized but the learning data was in form of protein sequence windows only.This study reports the GEARS model which is a novel machine leaning simulator (under development). The development phases are divided into various functional and reportable deliverable. This study will report following functional deliverables, conceptual overview of the simulator and its utilization scheme, bio-data preprCessing modules, MAPRes and neural network subsystems. Functional modules (MAPRes and Neural Network) reported here can be used to train a classifier based on the induction of data as well as significant biological rules capable to perform prediction. These reported modules will use to develop a novel prediction model with probably highest rate of accuracy and efficiency. In this study, a test dataset of phosphorylatedTyrosine (PTyr) modified by EGFR kinase were brought into account to validate the function and hypothesis of GEARS model. Not only the GEARS model but this test module too have the capacity to train data for the prediction of other modified residues. The availability of the data is the key factor for the training of these suggested modules.

MATERIALS AND METHODS

The GEARS modelGEARS, is a classifier evolution and learning manager that work on the bases of parallel and distributed computing. The structure of the GEARS is designed on layered and plug-in architecture. The layered architecture will have the capacity to handle the computational complexities of genetic algorithm to evolve respective models while plug-in architecture will provide the flexibility to handle different types of data servers. Software artifacts of GEARS system are grouped into three layers (data repository layer, windows application module layer and windows service module layer) (Fig. 1). On the bases of these layers, the GEARS model can be integrated into three basic modules which are further divided into some subsystems. The development of GEARS model is executing in C# (Fig. 2.). The overview of the main modules and subsystems of GEARS model are described below.

Windows application modules layerIn this module many software artifacts arecomposed in a group. These software artifacts are Classifier Evolution and Learning Manager (CEL Manager), Job Monitoring Console (JMC) and Project Registry Explorer (PRE).The CEL Manager will provide an interface to human user to create and configure GEARS project. In this module, after the defining the project, rest of the configurations such project registry and repositories will be created by GEARS itself. This application will also provide the interface to run the learning and evolution simulation. CEL Manager will communicate with configuration parameters like the software artifacts (Services) in Windows Service Modules Layer (WSML). Services in WSML will be responsible to configure and run the respective jobs. Data management of input and output data sources will be mange by Data Repository Layer (DRL) that will be based on plug-in architecture. The user required plug-ins will be select at the time of project creation and application will load the installed plug-ins. This will make the GEARS framework flexible enough to handle any type of data server.In next step, Job Monitoring Console application will provide the interface to illustrate the job status, progress of dot net and grid threads, learning curves etc. This application receives data streams from Activity Log Service (ALS) (Fig. 2).In last step, the Project Repository Explorer (PRE) will be utilize to browse the reports, configurations and architecture of evolved classifier candidate. This explorer will also provide the support to generate and compile the selected candidate as a dot net (DLL) component.

Windows service modules layerThis layer will consist of different Windows services which are the part of GEARS subsystems such as MAPRes Evolution Engine Service, ANN Evolution Engine Service, HMM Evolution Engine Service and Activity Log Service (Fig. 2). The idea behind splitting of GEARS application is to balance the computational load in distributed fashion. Each distribution is a grid application (except for Activity Log Service) and will be responsible to perform their specific tasks. Application Log Service which is an interface between other GEARS services, CEL Manager and Job Monitoring Console. This service will have the responsibility to entertain incoming data streams from other GEARS services and CEL Manager, writes in a log database and stream them to Job Monitoring Console application. The data streams will contain the information about the status and progress of each dot net and grid thread executing in other services.The MAPRes, ANN and HMM Evolution Engine Services are the examples of grid services and the development of these models are based on genetic algorithm. These Windows evolution engine services are responsible to evolve their respective classifier models e.g., MAPRes Evolution EngineServices will evolve MAPRes classification model. Each model will be executed as grid thread and to enhance the further computational efficiency the dot net threads will be executing within each candidate model.

Data repository layer (DRL)All Windows applications and services which are the members of GEARS system, will communicate with DRL with exception for ALS and JMC. Activity log database is not the part of DRL because log database can be viewed as temporary storage. On every execution of CEL Manager, activity log database will be cleaned. Whereas, data objects in DRL will persist in their states as long as project registry is not deleted by Project Registry Manager. In other words, each project will have its own unique DRL profile and will be treated as internal storage. As mentioned earlier that access to data objects in DRL will be performed through plug-in interface, means the GEARS implementa- tion is not bound to a specific arrangement of DRL data objects and their database servers.DRL management plug-ins is the implementation of DRL management interface library. MYSQL server based DRL management (MySQLDRL) plug-in will be implemented as default. Each data object in MYSQLDRL will be treated as a separate database, so that each data object can utilize maximum data storage. If required in any study to have different management strategy and/or database server, one can implement his/her own DRL management plug-in by extending interfaces defined in DRL management Interface Library.Data object in DRL are following: MAPRes Data Object, ANN Data Object, HMM Data object, Input Source Data Object, Global Genetic Algorithm (Global GA) Data Object, Project Configuration Log and Project Registry (Fig. 2).Data Objects assCiated to MAPRes, ANN and HMM are responsible to hold information on the respective model architecture and configurations, learning curves and error rates, training, validation and testing data sets, testing/validation reports for each candidate against its ModelID, Generation No. and initial conditions. Whereas, Global GA Data Object will be responsible to persist information on overall learning profile including learning curves and error rates of candidate models in a combinatorial way (ensemble mode), accommodating the new configurations of the data including size of peptides, training/validation/testing datasets.Several components of GEARS have beendeveloped and tested such as Project Creation Module, Main Console Module and Ensemble Learning Module. In current study, the justifications of relations behind the GEARS is performed by considering the assembled learning using ANN and MAPRes subsystems to evolve a model to predict the modification potential of PTyr data modified by EGFR kinase.

Test model to predict the modification potential ofPTyr dataA part of the proposed model was tested by developing a prediction-model for the prediction of the PTyr modified by EGFR kinase. The PTyr data based on EGFR kinase was obtained from Phospho.ELM. Data preprCessing and cleaning was performed by using Data Inconsistency modules of MAPRes in order to remove the inconsistencies in downloaded data from Phospho.ELM.After cleaning of data, the MAPRes model was utilized to mine the assCiation patterns from the resultant dataset. MAPRes mine these assCiation patterns at different support levels (5%,10%, 15%, 20% and 25%). The resultant dataset of PTyr were also take into account to make a peptide dataset of modified residues, the length of each peptide were consist of 21 amino acids (modified residue at 0 position and 10 amino acids at each side). This peptide dataset were utilized to train and validate the neural networks. The assCiation patterns mined by MAPRes were compared with each peptide of the peptide dataset to find the suitable match. The assCiation patterns and peptides which were matched, combine as a single record. To generate the binary form of the amino acids, sparse encoding was applied on the peptide dataset. For each amino acid a binary code of 21 digit was generated hence a 420 digit code represent a single peptide. The position "0" was not considered for sparse encoding as it is populated The biasness factor was added against each encoded peptide on the bases of MAPRes rules for effective learning of neural network. After the calculation of the biasness factor, the bootstrap sampling prCess was introduced to train and validate the peptide dataset. The training of the neural network was performed by varying the values of the sigmoidal lambda and root mean square (RMS) error. There were 21 experiments in total was performed by changing these values to identify the neural network configuration with optimized values, maximizing Mathews CorrelationCoefficient (MCC), sensitivity, specificity, accuracy and minimizing RMS error (Table I).

Table I.- Parameter alterations for training of ANNs.

Sr.###Sigmoidal Lambda###Acceptable RMS Error

1.###0.01###0.07

2.###0.18

3.###0.02###0.07

4.###0.075

5.###0.09

6.###0.1

7.###0.1

8.###0.12

9.###0.13

10.###0.03###0.05

11.###0.06

12.###0.07

13.###0.08

14.###0.1

15.###0.11

16.###0.12

17.###0.04###0.05

18.###0.05

19.###0.08

20.###0.1

21.###0.05###0.04

Table II.- AssCiation patterns mined by MAPRes.

Sr.###AssCiation rule###Support level

1.###less than E,-3greater than less than V,1greater than =greater than PP1###5

2.###less than G,-6greater than less than E,-3greater than =greater than PP###5

3.###less than G,-6greater than less than V,1greater than =greater than PP###5

4.###less than E,-3greater than =greater than PP###20

5.###less than G,-6greater than =greater than PP###20

6.###less than V,1greater than =greater than PP###24

7.###less than F,-4greater than =greater than NP2###5

9.###less than I,2greater than =greater than NP###5

10.###less than I,-7greater than =greater than NP###5

11.###less than K,-3greater than =greater than NP###5

12.###less than Q,8greater than =greater than NP###5

13.###less than R,5greater than =greater than NP###5

RESULTS AND DISCUSSION

The GEARS simulator tends to evolve amino acid sequence and assCiation rule. The classifiers

Table III.-###Input and output parameters for training of ANNs.

###Sr. No###Sigmoidal Lambda###Acceptable RMS Error###MCC###Sensitivity###Specificity###Accuracy

1.###0.03###0.11###0.538703###1###0.719697###0.764331

2.###0.04###0.1###0.538703###1###0.719697###0.764331

3.###0.04###0.08###0.545951###1###0.727273###0.770701

4.###0.03###0.12###0.545951###1###0.727273###0.770701

5.###0.03###0.08###0.553342###1###0.734849###0.77707

6.###0.03###0.1###0.553342###1###0.734849###0.77707

7.###0.03###0.07###0.592655###1###0.772727###0.808917

8.###0.02###0.13###0.627409###1###0.80303###0.834395

9.###0.02###0.1###0.646096###1###0.818182###0.847134

10.###0.04###0.05###0.655806###1###0.825758###0.853503

11.###0.04###0.05###0.655806###1###0.825758###0.853503

12.###0.03###0.06###0.67603###1###0.840909###0.866242

13.###0.02###0.12###0.732015###1###0.878788###0.898089

14.###0.02###0.09###0.770131###1###0.901515###0.917198

15.###0.02###0.07###0.770131###1###0.901515###0.917198

16.###0.02###0.1###0.783741###1###0.909091###0.923567

17.###0.05###0.04###0.797856###1###0.916667###0.929936

18.###0.03###0.05###0.797856###1###0.916667###0.929936

19.###0.02###0.075###0.797856###1###0.916667###0.929936

20.###0.01###0.18###0.800714###0.68###1###0.949045

21.###0.01###0.07###0.8436007###1###0.939394###0.9490446

are evolve in this method by utilizing genetic algorithm. The classifiers which are evolved as a jury of machine learning models are MAPRes, ANNs and HMMs. GEARS uses the concept of heterogeneous jury of aforementioned models. Every model have its own inherited strengths and different nature of learning e.g., ANNs are empirical and connectionist system, HMMs have statistical grounds whereas MAPRes offers a rule learning mechanism from unsupervised to supervised learning. This combinatorial scheme of three machine learning techniques in GEARS provides it the capability to accommodate complexities in protein data.In this specific case of study, MAPRes andneural networks which are the two basic proposed modules of GEARS methodology were utilized to perform sequence learning with rule inductions. AssCiation rules which were used for rule induction in the form of biasness factor were generated by MAPRes and the sequence learning was performed by using neural networks.In first step of this method, the assCiation rules were generated at various support levels by using MAPRes (Table II). At different support levels, some similar assCiation rules were found that were removed from the list and kept only one rule that have the maximum support level. After removing the similar rules from the list, the remaining assCiation rules were have only three support levels 5, 20 and 24. There were only 13 assCiation rules were left in the list and out of these total assCiation rules, 6 were found for positively PTyr (PP) and 7 for non-phosphorylated Tyr (NP). Among the 6 assCiation rules for PP, 3 were found at 5% support level, 2 were found at 20% support level and 1 at 24% support level. Whereas, in case of NP all assCiation rules were found at 5% support level. The assCiation patterns indicated that the Val at +1 position were found at maximum support level (24%), while, Glu at -3 position and Gly at -6 position have considerably high support level. It is interestingly noticed that all these three amino acids, found in surrounding of PP with maximum support level are non-polar.There were 21 experiments were performed to train the neural networks by changing the values of the sigmoidal lambda and root mean square (RMS) error (Table III). These trainings of the data are critical for the prediction of potential PTyr. As mentioned earlier sigmoidal Lambda and acceptable RMS were the training configurations, whereas the training performance of the neural networks was measured on the bases of following performance parameters such as Mathews correlation coefficient (MCC), sensitivity, specificity and accuracy. These experiments were conducted with 5 different values of sigmoidal lambda. Variation in root mean square (RMS) error was recorded during the learning phase. The value of sigmoid lambda ranges from0.01 to 0.04 and RMS error from 0.04 to 0.18 correspondingly set the value of accuracy. The experiment 17 to 21 have shown the highest values of the accuracy than the previous experiments (Table III). It is indicating that the training of the data was good in these experiments. The maximum learning of the data was done in last experiment with maximum value of accuracy. The finally selected experiment for prediction which has MCC= 0.8436007, Sensitivity = 1, Specificity = 0.939394and Accuracy = 0.9490446. This finally selected trained dataset will have the capacity to develop a reliable prediction model of potential PTyr residues.In conclusion, the development of new computational prediction models by considering the combinatorial approaches is the need of the hour to improve the accuracy of the prediction models. This study proposes the architecture of the GEARS model which is designed to combine the different data learning and data mining techniques. The GEARS model will have the capacity to predict the modified residues with better accuracy. A part of the GEARS model have been developed and results of this module have shown a good values of data learning or accuracy. This test module were utilized to train the data of PTyr but it also have the capacity to execute on other modified residues such as Ser, Thr, Lys and Gln. The extensive data available for phosphorylation of OH- group compels to investigate this particular study. The modification on NH3- group such as acetylation and methylation are equally important and control the various function of the proteins. In future, the data from other modification will also be considered for the training of developed module of the GEARS. ACKNOWLEDGEMENTS

Nasir-ud-din acknowledges financial support from Pakistan Academy of Sciences for amino group studies.

REFERENCES

AHMAD, I., HOESSLI, D.C., WALKER-NASIR, E., CHOUDHARY, M.I., RAFIK, S.M., SHAKOORI, A.R. AND NASIR-UD-DIN, 2006. Phosphorylation and glycosylation interplay: protein modifications at hydroxy amino acids and prediction of signaling functions of the human beta3 integrin family. J. cell. BiChem., 99: 706-718.AHMAD, I., HOESSLI, D.C., QAZI, W.M., KHURSHID, A., MAHMOOD, A., WALKER-NASIR, E., AHMAD, M., SHAKOORI, A.R. AND NASIR-UD-DIN, 2008b. MAPRes: An efficient method to analyze protein sequence around post translational modification sites. J. cell. BiChem., 104: 12201231.AHMAD, I., MEHMOOD, A., KHURSHID, A., QAZI, W.M., HOESSLI, D.C., WALKER-NASIR, E., SHAKOORI, A.R. AND NASIR-UD-DIN, 2009. Phosphoproteome sequence analysis and significance: Mining assCiation patterns around phosphorylation sites utilizing MAPRes. J. cell. BiChem., 108: 64-74.AHMAD, I., QAZI, W.M., KHURSHID, A., AHMAD, M., HOESSLI, D.C., WALKER-NASIR, E., CHOUDHARY, M.I., KHWAJA, I., SHAKOORI, A.R. AND NASIR-UD-DIN, 2008a. MAPRes: Mining assCiation rules among preferred amino acid residues in the vicinity of target amino acids in proteins. Proteomics, 8: 19541958.ATTWOOD, T., 2000. The quest to deduce protein function from sequence: the role of pattern databases. Int. J. BiChem. Cell Biol., 32:139-155.BERLOT, S., AISSAOUI, Z., PAVON-DJAVID, G., BELLENEY, J., JOZEFOWICZ, M., HELARY, G. AND MIGONNEY, V., 2002. Biomimetic poly(methyl methacrylate)-based terpolymers: modulation of bacterial adhesion effect. Biomacromolecules, 3: 63-68.BLOM, N., SICHERITZ-PONTEN, T., GUPTA, R., GAMMELTOFT, S. AND BRUNAK, S., 2004. Prediction of postTranslational glycosylation and phosphorylation of proteins from the amino acid sequence. Proteomics, 4:1633-1649.BORK, P., DANSEKAR, T., DIAZ-LAZCOZ, Y., EISENHABER, F., HUYNEN, M. AND YUAN, Y.,1998. Predicting function: from genes to genome and back. J. mol. Biol., 283:707-725.BOYNTON, J.C.R., HEINEGARD, D. AND BARRY, F.,2001. Post translational modifications to human bone sialoprotein determined by mass spectrometry. BiChemistry, 40:12983-12991.BUTT, A., KALEEM, A., IQBAL, Z., WALKER-NASIR, E., SALEEM, M., SHAKOORI, A.R. AND NASIR-UD- DIN., 2011. Functional regulation of DNA binding of FOXO1 by post translational modifications: In silico study. Pakistan J. Zool., 43: 1167-1175.CHRISTLET, T.H.T. AND VELURAJA, K., 2001. Database analysis of o-glycosylation sites in proteins. Biophys. J.,80: 952-960.FUJII, K., ZHU, G., LIU, Y., HALLAM, J., CHEN, L., HERRERO, J. AND SHAW, S., 2004. Kinase peptide specificity: improved determination and relevance to protein phosphorylation. PrC. natl. Acad. Sci., USA,101: 13744-13749.GUPTA, R., JUNG, E., GOOLEY, A.A., WILLIAMS, K.L., BRUNAK, S. AND HANSEN, J., 1999. Scanning the available Dictyostelium discoideum proteome for O- linked GlcNAc glycosylation sites using neural networks. Glycobiology, 9: 1009-1022.IQBAL, Z., HOESSLI, D.C., KALEEM, A., MUNIR, J., SALEEM, M., AFZAL, I., SHAKOORI, A.R. AND NASIR-UD-DIN, 2013. Influence of the sequence environment and properties of neighboring amino acids on amino-acetylation: relevance for structure-function analysis. J. cell. BiChem., 114: 874-887.JEFFERY, C.J., 1999. Moonlighting proteins. Trends biChem..Sci., 24:8-11.KHWAJA, T.A., WAJAHAT, T., AHMAD, I., HOESSLI, D.C., WALKER-NASIR, E., KALEEM, A., QAZI, W.M., SHAKOORI, A.R. AND NASIR-UD-DIN,2008. In silico modulation of apoptotic Bcl-2 proteins by mistletoe lectin-1: functional consequences of protein modifications. J. cell. BiChem., 103: 479491.LEE, T., HUANG, H., HUNG, J., HUANG, H., YANG, Y.AND WANG, T., 2006 dbPTM: An information repository of protein post translational modification.Nucl. Acids Res., 34: D622--D627.LI, J., BIGELOW, D.J. AND SQUIER, T.C., 2004.Conformational changes within the cytosolic portion of phospholamban upon release of Ca-ATPase inhibition. BiChemistry, 43:3870-3879.QAZI, W.M., SHAKOORI, A.R. AND NASIR-UD-DIN.,2006. Rules required for phosphorylating tyrosine using as a model kinase EGFR utilizing AME. In: Abstracts of International Symposium on Nano Chemistry: Chemistry, BiChemistry, Molecular Biology and Bioinformatics of Enzymes, pp. 10.SCHIZA, V., MOLINA-SERRANO, D., KYRIAKOU, D., HADJIANTONIOU, A. AND KIRMIZIS, A., 2013. N- alphaTerminal acetylation of histone H4 regulates arginine methylation and ribosomal DNA silencing. PLoS. Genet., 9: e1003805.VARKI, A. AND KORNFELD, S., 1980. Structural studies of phosphorylated high mannoseType oligosaccharides. J biol. Chem., 255:10847-10858.WONG, Y., LEE, T., LIANG, H., HUANG, C., WANG, T., YANG. Y., CHU, C., HUANG, H., KO, M. AND HWANG, J., 2007. KinasePhos2.0: A web server for identifying protein kinase-specific phosphorylation sites based on sequences and coupling patterns. Nucl. Acids Res., 35:W588-594.YUAN, Z.Q., FELDMAN, R.I., BUSMAN, G.E., COPPOLA, D., NICOSIA, S.V. AND CHENG, J.Q., 2003. AKT2 inhibition of capsulation-induced JNK/p38 and box activation by phosphorylation of ASK1: implication of AKT2 in chemo resistance. J. biol. Chem., 278:23432-23440.
COPYRIGHT 2014 Asianet-Pakistan
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2014 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Qazi, Wajahat M.; Iqbal, Zeeshan; Khan, M. Saleem; Rehan, Muhammad; Munir, Jawaria; Shakoori, Abdul
Publication:Pakistan Journal of Zoology
Article Type:Report
Geographic Code:9PAKI
Date:Apr 30, 2014
Words:4337
Previous Article:Prevalence and Hematology of Tick Borne Hemoparasitic Diseases in Equines in and Around Lahore.
Next Article:Effect of Multi-Enzymes Supplementation on Growth Performance of Broiler.
Topics:

Terms of use | Privacy policy | Copyright © 2022 Farlex, Inc. | Feedback | For webmasters |