About HSLpred

 

The data set
Human proteins whose subcellular locations were annotated has been extracted from special release of SWISSPROT database. All the proteins which were residing in more then one location, annotated as "isoforms", "potential", "probable", "by similarity" were filtered out. Further sequence redundancy was reduced such that none had >90% sequence identity to any other. Overall we were remained with 3780 proteins out of 10777, which were divided into 11 groups according to their subcellular localization as shown in the table. However, amount of data available for some gropus was too small for statistical analysis to be performed. Final data set consisted of a total of 3532 human proteins (840 CYTOPLASMIC, 315 MITOCHONDRIAL, 858 NUCLEAR, 1519 PLASMA MEMBRANE).

Subcellular localization Number of proteins
Cytoplasm  840
Mitochondria 315
Nuclear 858
Plasma Membrane 1519
Total 3532

Support Vector Machines (SVM)
In the present study, we have used SVMlight to predict the subcellular localization of human proteins. The software enables the users to define a number of parameters and also allows a choice of inbuilt kernel functions, including linear, RBF and polynomial. The prediction of subcellular localization is a multi-class classification. Here, the class number was equal to four for human proteins. The ith SVM was trained with all the samples in the ith class with positive labels and all other samples with negative labels. In this way, four SVMs were constructed for subcellular localization of proteins to nuclear, mitochondrial, cytoplasmic, and plasma membrane. An unknown sample was classified into the class that corresponded to the SVM with highest output score.
 
Evaluation of HSLpred
The performance modules constructed in this study has been evaluated using 5-fold cross-validation technique. In this technique, the relevant dataset was partitioned randomly into five equally sized sets. The training and testing was carried out five times, each time using one distinct set for testing and the remaining four sets for training. For evaluating the performance of various modules, accuracy and Matthew’s correlation coefficient (MCC) were calculated using the following equations:

 

where, x can be any subcellular location (cytoplasmic, mitochondrial, nuclear, or plasma membrane), exp(x) is the number of sequences observed in location x, p(x) is the number of correctly predicted sequences of location x, n(x) is the number of correctly predicted sequences not of location x, u(x) is the number of under predicted sequences and o(x) is the number of over-predicted sequences.


Prediction Approaches

Different approaches have been used for the subcellular localization of human proteins. These approaches are based on proteins features such as:
 
Amino acid composition is the fraction of each amino acid in a protein. The calculation of amino acid composition  generates the 20 dimensional input vectors which were used  to train four types of SVM models for the four types of  subcellular localizations. The composition based SVM module (kernel=RBF, g= 300, C=2, j=1) was able to predict with overall accuracy of 76.6%. the results obtained are shown below:

 

Subcellular localization  Accuracy (%) MCC
Cytoplasm 63.5 0.52
Mitochondria 46.1 0.52
Nuclear 76.2 0.67
Plasma Membrane 90.3 0.78
 

Dipeptide Composition was used to encapsulate the global information about each protein sequence, which gives a fixed pattern length of 400 (20 X 20). This representation encompassed the information about amino acid composition along local order of amino acid. In the case of 1-2dipeptide SVM module the best results were achieved with the RBF kernel (g=50, C=6, j=1). The SVM module was predicted with 77.8% overall accuracy which was nearly 1% better then amino acid composition based SVM module. The results obtained after 5-fold cross-validation for 1-2diepptide composition are shown below:

 

Subcellular localization  Accuracy (%) MCC
Cytoplasm 58.3 0.52
Mitochondria 48.3 0.52
Nuclear 80.2 0.71
Plasma Membrane 93.4 0.80
 
PSIBLAST a module was designed in which a query sequence was searched against a non redundant database available at NCBI using PSI-BLAST. In the present study, PS_-BLAST was used instead of normal standard BLAST because it has the capability to detect remote homologies. It carries out an iterative search in which sequences found in one round were used to build score model for the next round of searching. Three iterations of PSI-BLAST were carried out at a cut-off E-value of 0.001. This module could predict any of the four localizations depending upon the similarity of the query protein to the proteins in the dataset. The module would return "unknown subcellular localization" if no significant similarity was obtained. The accuracy obtained for four types of subcellular localization are shown below
 
Subcellular localization  Accuracy (%)
Cytoplasm 56.9
Mitochondria 40.6
Nuclear 68.2
Plasma Membrane 92.0

HYBRID SVM module

This module uses the complete information about the protein that is amino acid composition, dipeptide composition and evolutionary information of PS-BLAST output. SVM was provided with an output vector 425 dimensions that consisted of 20 for amino acid composition, 400 for dipeptide composition, five for PSI-BLAST output. The performance of this module was better then any other individual feature based module. This hybrid module with the RBF kernel (g=50, C=2, j=1) was able to achieve overall 84.9% accuracy. The results obtained for four types of subcellular localization are shown below:

Subcellular localization  Accuracy (%) MCC
Cytoplasm 75.4 0.67
Mitochondria 69.8 0.68
Nuclear 82.4 0.79
Plasma Membrane 94.8 0.89

Other SVM modules

We have also constructed other various types of dipeptide SVM module (1-3, 1-4 and 1-5) and hybrid modules . The accuracy obtained for the SVM modules are shown in the graph below

 

 where

1 is 1-2dipeptide composition (D)

2 is 1-3dipeptide composition (E)

3 is 1-4dipeptide composition (F)

4 is 1-5dipeptide composition (G)

5 is hybrid2 SVM module (H=D+E+F+G)

6 is hybrid3 SVM module (Amino acid +H+ PSI-BLAST)

7 is a cascade SVM

RELIABILITY INDEX

The reliability index (RI) is a commonly used measure of prediction that provides confidence about a prediction to the users. The RI assignment is a useful indication of the level of certainty in the prediction for a particular sequence. The RI was assigned according to the difference between the highest and second highest SVM output scores. We have also computed the reliability score of our prediction method based on the hybrid approach using the following equation