Khaled Ben Islam

Identification of post-translational modifications (PTM) is significant in the study of computational proteomics, cell biology, pathogenesis, and drug development due to its role in many bio-molecular mechanisms. Computational methods for predicting multiple PTM at the same lysine residues, often referred to as K-PTM, is still evolving. This paper presents a novel computational tool, abbreviated as predML-Site, for predicting K-PTM, such as acetylation, crotonylation, methylation, succinylation from an uncategorized peptide sample involving single, multiple, or no modification. For informative feature representation, multiple sequence encoding schemes, such as the sequence-coupling, binary encoding, k-spaced amino acid pairs, amino acid factor have been used with ANOVA and incremental feature selection. As a core predictor, a cost-sensitive SVM classifier has been adopted which effectively mitigates the effect of class-label imbalance in the dataset. predML-Site predicts multi-label PTM sites with 84.18% accuracy using the top 91 features. It has also achieved 85.34% aiming and 86.58% coverage rate which are much better than the existing state-of-the-art predictors on the same rigorous validation test. This performance indicates that predML-Site can be used as a supportive tool for further K-PTM study. For the convenience of the experimental scientists, predML-Site has been deployed as a user-friendly web-server at http://103.99.176.239/predML-Site.

Background: Post-translational modification is one of the bio-molecular mechanisms in living organisms, which incorporate functional diversity in proteins as well as regulate cellular processes. Transformation of arginine residue to citrulline in protein is such a modification. Objective: Our objective is to identify citrullinated arginine residue sites quickly and accurately. Methods: In this study, a novel computational tool, abbreviated as predCitru-Site, has been developed to predict citrullination sites. This technique effectively has incorporated the sequencecoupling effect of surrounding amino acids of arginine residues as well as optimizes skewed training citrullination dataset for prediction quality improvement. The performance of predCitru- Site has been measured from the average of 5 complete runs of the 10-fold cross-validation test to comply with existing tools. Results and Conclusion: predCitru-Site has achieved 97.6% sensitivity, 98.9% specificity, and overall accuracy of 98.5%. With Matthew’s correlation coefficient of 0.967, it has also shown an area under the receiver operator characteristics curve of 0.997. Compared with existing tools, predCitru-Site significantly outperforms on the same benchmark dataset. It also shows significant improvement in the case of independent tests in all performance metrics (around 50% higher in AUC). These results suggest that our method is promising and can be used as a complementary technique for fast exploration of citrullination in arginine residue. A user-friendly web server has also been deployed at http://research.ru.ac.bd/predCitru-Site/ for the convenience of experimental scientists.

Finding an optimal subspace of bands that has the most expressive power for classifying hyperspectral image has been very challenging task due to its insufficient number of training pixels with respect to large number of bands. Feature reduction is considered a promising solution in this type of task. However, it is very hard to select an optimal feature reduction technique which is effective as well as computationally efficient in case of hyperspectral image classification. Moreover, it becomes challenging when the number of training pixels of a class is not sufficient. In this paper, we have rigorously studied some feature selection techniques for reducing spectral dimension by considering all the classes in hyperspectral image on a benchmark data set. We have projected that this study will be very supportive for further study on band selection and hyperspectral image classification.

Prediction of protein subcellular localization is the most challenging field for the researchers because of its importance in different branch of molecular biology and drug discovery. Last two decades, a large number of machine learning approaches have been tested into sequence based features for the prediction of subcellular localization. Single features like amino acid composition (AAC), pseudo amino acid composition (PseAAC) and physiochemical property model (PPM)) contain insufficient information due to their single perspectives. To overcome this problem, the main contribution of our work is to propose two feature fusion representations AACPPM and PAACPPM which can be fused PPM with AAC and PseAAC respectively. Support Vector Machine (SVM) is applied as a classifier on to both single and fused feature representations of Gram-positive bacterial dataset. The actual accuracy of AACPPM is 72.4% which is 2% higher than single feature representations and 6% higher than X. Qu et al [1]. The locative accuracy of both AACPPM and PAACPPM is 73.2% which is also 2% higher than single feature representations.

Cloud computing is one of the buzzwords of recent time. Cloud computing can be defined as a new style of computing in which dynamically scalable and often virtualized resources are provided as a service over the Internet. Unfortunately, though the Banking sector which works as a financial intermediary and one of the major industries for monetary activities is still deprived of the advantages of cloud. The major reason behind this is the lack of confidentiality and security. As a solution, we are proposing private cloud architecture for banking sector. So their confidentiality will be maintained alongside their data will be safe and sound. We are proposing a simple yet efficient load allocation algorithm to match the homogeneous nature of the banking tasks. We are also proposing a modular banking system. This solution is particularly helpful for developing countries, rural and remote places where setting banking infrastructure is not possible or feasible.

In recent years, a large number of proteins of different organisms have been discovered but due to high experimental cost and uncertain time boundary, yet it is not possible to find out all of the functionalities of those proteins. With the recent advent of huge protein-protein interactions, it becomes an opportunity to computationally predict a protein’s functionality based on its interacting partners. In this work, we mainly try to find out a way by which we can predict functionality of a target protein with low computational complexity. We propose a simple approach for protein function prediction based on Classical Neighbor Counting method. We also investigate the functional dependency of a protein to its direct neighbors in the interaction network. We find that when majority of its interacting partners have more experimentally known annotation, then more accurately we can predict a protein’s functionality using Neighbor Counting technique.

In this paper, we have analysed the comparative performance of AODV, DSDV and DSR routing protocols in different wireless sensor network scenarios (i.e. scenarios where number of nodes changes or mobility of nodes changes). The performance matrix includes Packet Delivery Ratio, Throughput, End to End Delay, and Normalized Routing Load. We mainly try to address the behaviour of the protocols in different scenarios. Simulation results show that different routing protocol performs well in different scenarios and good for specific performance metrics. For example, DSDV perform better in the high density networks or the network with strict requirement on time whereas DSR performs well in smaller network. AODV is more adaptable in the networks with high throughputs and preferable for low loss rate environment.