| Research Article |
Open Access |
|
| SVM Model for Amino Acid Composition Based Prediction of Mycobacterium
tuberculosis |
| Lakshmi Pillai1*, Bhasker Pant1, Usha chauhan2 and KR Pardasani3 |
| 1Phd Student, Department of Mathematics, Maulana Azad National Institute of Technology |
| 2Assistant Professor, Department of Mathematics, Maulana Azad National Institute of Technology |
| 3Professor, Department of Mathematics, Maulana Azad National Institute of Technology |
| *Corresponding author: |
Dr. Lakshmi Pillai
Department of Mathematiics
Maulana
Azad National Institute of Technology
Bhopal, India
Tel: +91 (0)7828013946
E-mail: lakshmilster@gmail.com |
|
| |
| Received May 12, 2011; Accepted July 30, 2011; Published July 31, 2011 |
| |
| Citation: Pillai L, Pant B, Chauhan U, Pardasani KR (2011) SVM Model for Amino
Acid Composition Based Prediction of Mycobacterium tuberculosis. J Comput Sci
Syst Biol 4: 047-049. doi:10.4172/jcsb.1000075 |
| |
| Copyright: © 2011 Pillai L, et al. This is an open-access article distributed under
the terms of the Creative Commons Attribution License,which permits unrestricted
use, distribution, and reproduction in any medium, provided the original author and
source are credited. |
| |
| Abstract |
| |
| The Tuberculosis is the classical human mycobacterial disease, caused by Mycobacterium tuberculosis.
The disease primarily affect the lung and causes pulmonary tuberculosis, as well as affect intestine, bone, joints,
meninges, lymph nodes, skin and other tissue of the body, causing extra pulmonary tuberculosis. Thus there arises
the need to understand the relationships among various parameters of these proteins for prediction of their classes,
structures and functionality. The computational approaches for prediction of their classes are fast and economical
therefore can be used to complement the existing wet lab techniques. Realizing their importance, in this paper an
attempt has been made to correlate them with their amino acid composition and predict them with fair accuracy. This
is a novel method where Mycobacterium Tuberculosis has been classified on the basis of amino acid composition
using Support Vector Machine. The SVM has been implemented using SVM Light package [1,2]. The method
discriminates different strains of Mycobacterium Tuberculosis. The performance of the method was evaluated using
10-fold cross-validation where accuracy of 100% was obtained. |
| |
| Keywords |
| |
| xtra pulmonary Tuberculosis; Support vector machine;
Amino acid composition; Kernel functions; Granuloma; Macrophages;
Necrosis; Binary classifier; Cytotoxic T cells; Supervised machine
learning; Matthews correlation coefficient |
| |
| Introduction |
| |
| The German scientist (Robert Koch) announced that he had
cultured the causative agent from human TB lesions and designated as
"Bacillus of Tuberculosis". Mycobacterium, the genus of Actinobacteria,
given its own family of mycobacteriaceae includes certain species. |
| |
| About 90% of those infected with Mycobacterium
tuberculosis have asymptomatic, latent TB infection (sometimes called
LTBI), with only a 10% lifetime chance that a latent infection will
progress to TB disease. However, if untreated, the death rate for these
active TB cases is more than 50%. |
| |
| TB infection begins when the mycobacteria reach the pulmonary
alveoli, where they invade and replicate with the endosomes of alveolar
macrophages. The primary site of infection in the lungs is called
the Ghon focus, and is generally located in either the upper part of the
lower lobe, or the lower part of the upper lobe. Bacteria are picked up
by dendritic cells, which do not allow replication, although these cells
can transport the bacilli to local (mediastinal) lymph nodes. Further
spread is through the bloodstream to other tissues and organs where
secondary TB lesions can develop in other parts of the lung (particularly
the apex of the upper lobes), peripheral lymph nodes, kidneys, brain,
and bone. All parts of the body can be affected by the disease, though
it rarely affects the heart, skeletal muscles, pancreas and thyroid.
Tuberculosis is classified as one of the granulomatous inflammatory
conditions. Macrophages, T-lymphocytes, B-lymphocytes and
fibroblasts are among the cells that aggregate to form agranuloma, with
lymphocytes surrounding the infected macrophages. The granuloma
functions not only to prevent dissemination of the mycobacteria, but also
provides a local environment for communication of cells of the immune
system. Within the granuloma, T lymphocytes secrete cytokines such
as interferon gamma, which activates macrophages to destroy the bacteria with which they are infected. Cytotoxic T cells can also directly
kill infected cells, by secreting perforin and granulysin [3,4]. |
| |
| Importantly, bacteria are not always eliminated within the
granuloma, but can become dormant, resulting in a latent infection.
Another feature of the granulomas of human tuberculosis is the
development of cell death, also called necrosis, in the center of tubercles.
To the naked eye this has the texture of soft white cheese and was
termed caseous necrosis. |
| |
| If TB bacteria gain entry to the bloodstream from an area of
damaged tissue they spread through the body and set up many foci of
infection, all appearing as tiny white tubercles in the tissues. This severe
form of TB disease is most common in infants and the elderly and is
called miliary tuberculosis. Patients with this disseminated TB have a
fatality rate of approximately 20%, even with intensive treatment. |
| |
| In many patients the infection waxes and wanes. Tissue destruction
and necrosis are balanced by healing and fibrosis. Affected tissue is
replaced by scarring and cavities filled with cheese-like white necrotic
material. During active disease, some of these cavities are joined to the
air passages bronchi and this material can be coughed up. It contains
living bacteria and can therefore pass on infection. Treatment with
appropriate antibiotics kills bacteria and allows healing to take place.
Upon cure, affected areas are eventually replaced by scar tissue. |
| |
| Currently, efforts are underway to develop new therapeutic
agents and elucidation of metabolic pathway associated with diseases [5]. Moreover, the mere understanding of different strains of
Mycobacterium will assist in finding novel drug target with minimum
side effects. The experimental attempts are reported in the literature
for functional classification of Mycobacterium Tuberculosis. But no
computational technique is available in the literature for classification
of Mycobacterium tuberculosis based on other parameters like
dipeptide composition, amino acid composition and physiochemical
properties. Since the experimental identifications of them are labor
and cost-intensive task, the computational biology can provide a better
alternative to develop a method for classifying different strains of
Mycobacterium Tuberculosis. |
| |
| In view of the above an attempt has been made in this paper to
develop a computational approach for predicting and classifying two
types of Tuberculosis strains i.e. Mycobacterium Tuberculosis and Non
Mycobacterium Tuberculosis. This is a binary classification method
where the Tuberculosis can be discriminated as Mycobacterium
Tuberculosis and Non Mycobacterium Tuberculosis [6]. It has been
shown in past that SVM is an elegant technique for the classification of
biological data. Here SVM model has been developed for amino acid
composition based prediction identification and classification of MTB
and Non Mycobacterium TB. |
| |
| This paper is a step in the direction where machine learning and
computational biology techniques can be used to complement existing
wet lab techniques [6,7]. |
| |
| Materials and Methods |
| |
| Data set |
| |
| To achieve our goal and develop our methodology we obtained the
dataset from Swissprot/Uniprot databank of Expasy server (12). The
following two data sets were used. |
| |
| Dataset1: It consisted of all Mycobacterium Tuberculosis proteins.
All the entries marked as fragments were not included in the dataset.
The total instances were 28,200. The final dataset consisted of 28,200
sequences belonging to Mycobacterium TB strains i.e. H37RA 4002
sequences, CDC1551 4197 sequences, F11 3905 sequences, H37RV
8021 sequences and KZN- 1435 4026 sequences [8,9]. |
| |
| Dataset2: To validate our methodology proteins belonging to some
other class were taken into consideration. They were treated as negative
instances. |
| |
| For training dataset we consider 25,000 sequences belonging to
different strains while remaining 3,200 sequences were used to prepare
the test dataset. |
| |
| Support vector machine (Binary classification) |
| |
| SVM is a supervised machine learning method which is based
on the statistical learning theory. When used as a binary classifier, an
SVM will construct a hyperplane, which acts as the decision surface
between the two classes. This is achieved by maximizing the margin of
separation between the hyperplane and those points nearest to it. The
SVMs were implemented using freely downloadable software, SVM
light [1,2]. In this software there is a facility to define parameters and
choose among various inbuilt kernals. They can be radial basis function
(RBF) or a polynomial kernel (of given degree), linear, sigmoid. |
| |
| SVM software; SVM light |
| |
| Simulations were performed using SVM light version 6.02 (a freely
available software package) [8,9]. For our study RBF Kernel was found to be the best. The SVM training was carried out by the optimization of
the value of the regularization parameter and the value of RBF kernel
parameter. |
| |
| Amino acid composition |
| |
| Previously, this parameter has been used for predicting the
subcellular localization of proteins [12,13]. The amino acid composition
is the fraction of each amino acid type within a protein. The fractions of
all 20 natural amino acids were calculated by using Equation 1, |
| |
| Fraction of amino acid i (where i can be any amino acid) |
| |
 |
| |
| Evaluation of Performance: |
| |
| The performance of our classifier was judged by 10 fold cross
validation. The SVM Light provides a parameter selection tool using
the RBF kernel: cross validation via grid search. A grid search was
performed on C and Gamma using an inbuilt module of SVM Light
tools as shown in Figure 1. Here pairs of C and Gamma are tried and
the one with the best cross validation accuracy is picked. On using
the values of C=0.5 and Gamma=0.5 obtained through grid search an
accuracy of 100% was obtained. |
| |
| Prediction system assessment |
| |
| True positives (TP) and true negatives (TN) were identified as the
positive and negative samples, respectively. False positives (FP) were
negative samples identified as positive. False negatives (FN) were
positive samples identified as negative. The prediction performance
was tested with sensitivity (TP/ (TP+FN)), specificity (TN/ (TN+FP)),
overall accuracy (Q2), and the Matthews correlation coefficient (MCC).
The accuracy and the MCC for each subfamily of Mycobacterium
tuberculosis, was calculated as described by Hua and Sun [9] and
shown below in equation 2 and 3. |
| |
 |
| |
 |
| |
| Results and Discussion |
| |
| All strains of Mycobacterium Tuberculosis have been implicated in
various diseases. Realizing their involvement in disease we have chosen
these strains for our study. |
| |
| The results obtained here will be helpful in differentiating
between different strains of Mycobacterium Tuberculosis. A new
protein discovered can be shown as belonging to any of the classified
Mycobacterium strain. This model can also be an important tool to
understand the differences between different strains hence a step
towards assisting various wet lab techniques in devising novel drugs
and therapeutic agents against these strains. The correlation of different
strains with their amino acid composition explored here can be useful
to obtain better insight about these strains. |
| |
| The overall accuracy and MCC of the amino acid composition-based
classifier [14] for classifying the different strains of Mycobacterium
Tuberculosis was 100%. It proved that strains can be correlated with
amino acid composition and can be easily distinguished on this basis. |
|
| |
|
Figure 1: Coarse Grid Search on C = 2-5, 2-4 . 210 and Gamma = 25, 24
. 2 - 10. |
|
| |
| Conclusion |
| |
| The SVM model developed here is computationally efficient and
effective in predicting and classifying the Mycobacterium Tuberculosis.
This is evident from the accuracy (100%) in the results. Further the
amino acid composition contains very significant information for
discriminating the classes of above proteins. |
| |
| This model can be used to analyze other strains, such as entire
proteomics data. Such type of prediction systems can be very useful for
understanding the above proteases in a better way so as in conclusion, a
novel method for classifying Mycobacterium Tuberculosis is presented.
This method will nicely complement the existing wet lab methods. It
will assist in assigning the correct class to which these proteins belong.
The prediction method presented here may be useful for the annotation
of the piled-up proteomic data. |
| |
| The author awaits discovery of more of these proteins in the future
so that accuracy of the prediction model can be increased further and a
server developed for public use. |
| |
| Key points |
| |
| . The SVM model developed here is computationally efficient
and effective in predicting and classifying the Mycobacterium
Tuberculosis with accuracy rate of 100%. |
| |
| . This model can be used to analyze other strains, such as entire
proteomics data of Mycobacterium Tuberculosis. |
| |
| . All strains of Mycobacterium Tuberculosis have been
implicated in various diseases. Realizing their involvement in
disease we have chosen these strains for our study. |
| |
| . It will assist in assigning the correct class to which the proteins
belong and thus will nicely complement the existing wet lab
methods. |
| |
| . The correlation of different strains with their amino acid
composition explored here can be useful to obtain better
insight about these strains. |
| |
| Acknowledgements |
| |
| The authors are highly thankful to the department of Biotechnology, Delhi,
India and M.P. Council of Science and Technology M.P., Bhopal, India for providing
support in the form of Bioinformatics Infrastructure facility to carry out this work. |
| |
|
| References |
| |
- Cai YD, Liu XJ, Xu XB, Chou KC (2002) Support vector machines for prediction
of protein subcellular location by incorporating quasi-sequence-order effect. J
Cell Biochem 84: 343-348.
- Cai YD, Pong-Wong R, Feng K, Jen JCH, Chou KC (2004) Application of SVM
to predict membrane protein types. J Theor Biol 226: 373-376.
- Cai YD, Zhou GP, Jen CH, Lin SL, Chou KC (2004) Identify catalytic triads
of serine hydrolases by support vector machines. J Theor Biol 228: 551-557.
- Joachims T (2002) SVM Light: a library for support vector machines, software.
- Holmgren NB, Millman I, Youmans GP (1954) Studies on the metabolism of
Mycobacterium tuberculosis. VI. The effect of Krebs' tricarboxylic acid cycle
intermediates and precursors on the growth and respiration of Mycobacterium
tuberculosis. J Bacteriol 68: 405-410.
- Jamshidi N, Palsson BO (2007) Investigating the metabolic capabilities of
Mycobacterium tuberculosis H37Rv using the in silico strain iNJ661 and
proposing alternative drug targets BMC Syst Biol 1: 26.
- ExPASy Proteomics Server.
- Hua S, Sun Z (2001) Support vector machine approach for protein subcellular
localization prediction. Bioinformatics 17: 721-728.
- Cai YD, Zhou GP, Jen CH, Lin SL, Chou KC (2004) Identify catalytic triads
of serine hydrolases by support vector machines. J Theor Biol 228: 551-557.
- Joachims T (2001) SVM Light: a library for support vector machines.
- Schneider E, Moore M, Castro KG (2005) Epidemiology of tuberculosis in the
United States. Clin chest med 26: 183-195.
- Chou KC, Cai YD (2002) Using functional domain composition and support
vector machines for prediction of protein subcellular location. J Biol Chem 277:
45765-45769.
- Bhasin M, Raghava GP (2004) Classification of Nuclear Receptors Based on
Amino Acid Composition and Dipeptide Composition. J Biol chem 279: 23262-
23266.
- Blobel CP (1999) Metalloprotease-disintegrins: modular proteins capable of
promoting cell-cell interactions and triggering signals by protein-ectodomain
shedding. Cell 90: 589.
- Cosma CL, Sherman DR, Ramakrishnan L (2003) The secret lives of the
pathogenic mycobacteria. Annu Rev Microbiol 57: 641-676.
|
| |
| |
