Hafeez Ur Rehman

Using Boolean networks to model post-transcriptional regulation in gene regulatory networks

Abstract Gene regulatory networks (GRNs) model some of the mechanisms that regulate gene expression. Among the computational approaches available to model and study GNRs, Boolean network (BN) emerged as very successful to better understand both the structural and dynamical properties of GRNs. Nevertheless, the most widely used models based on BNs do not include any post-transcriptional regulation mechanism. Since miRNAs have been proved to play an important regulatory role, in this paper we show how the post-transcriptional regulation mechanism mediated by miRNAs has been included in an enhanced BN-based model. We resort to the miR-7 in two Drosophila cell fate determination networks to verify the effectiveness of miRNAs modeling in BNs, by implementing it in our tool for the analysis of Boolean networks.

Combining homolog and motif similarity data with Gene Ontology relationships for protein function prediction

Uncharacterized proteins pose a challenge not just to functional genomics, but also to biology in general. The knowledge of biochemical functions of such proteins is very critical for designing efficient therapeutic techniques. The bottleneck in hypothetical proteins annotation is the difficulty in collecting and aggregating enough biological information about the protein itself. In this paper, we propose and evaluate a protein annotation technique that aggregates different biological information conserved across many hypothetical proteins. To enhance the performance and to increase the prediction accuracy, we incorporate term specific relationships based on Gene Ontology (GO). Our method combines PPI (Protein Protein Interactions) data, protein motifs information, protein sequence similarity and protein homology data, with a context similarity measure based on Gene Ontology, to accurately infer functional information for unannotated proteins. We apply our method on Saccharomyces Cerevisiae species proteins. The aggregation of different sources of evidence with GO relationships increases the precision and accuracy of prediction compared to other methods reported in literature. We predicted with a precision and accuracy of 100% for more than half proteins of the input set and with an overall 81.35% precision and 80.04% accuracy.

A three-way approach for protein function classification

The knowledge of protein functions plays an essential role in understanding biological cells and has a significant impact on human life in areas such as personalized medicine, better crops and improved therapeutic interventions. Due to expense and inherent difficulty of biological experiments, intelligent methods are generally relied upon for automatic assignment of functions to proteins. The technological advancements in the field of biology are improving our understanding of biological processes and are regularly resulting in new features and characteristics that better describe the role of proteins. It is inevitable to neglect and overlook these anticipated features in designing more effective classification techniques. A key issue in this context, that is not being sufficiently addressed, is how to build effective classification models and approaches for protein function prediction by incorporating and taking advantage from the ever evolving biological information. In this article, we propose a three-way decision making approach which provides provisions for seeking and incorporating future information. We considered probabilistic rough sets based models such as Game-Theoretic Rough Sets (GTRS) and Information-Theoretic Rough Sets (ITRS) for inducing three-way decisions. An architecture of protein functions classification with probabilistic rough sets based three-way decisions is proposed and explained. Experiments are carried out on Saccharomyces cerevisiae species dataset obtained from Uniprot database with the corresponding functional classes extracted from the Gene Ontology (GO) database. The results indicate that as the level of biological information increases, the number of deferred cases are reduced while maintaining similar level of accuracy.

A comparative study of graphic symbol recognition methods

From the very beginning of written scripts, contents of documents generally comprise of text, images, figures, graphs and graphic symbols. A graphic recognition system involves representation of graphic symbols, description of features extracted from the symbol and classification of the unknown symbols. Due to the wide range of symbols, no generalize technique is available that can recognize the symbol for all the application domains. this paper, we present an overview of the many models and methodologies available to symbol recognition for representation, description and classification. We provide a general survey of symbol recognition process, beginning with the basic definition of symbol, which is further classified into their types based on application areas. distinctive part of the survey is categorization of different symbol recognition methods into four categories i.e. statistical, structural, syntactical and hybrid methods, which is aimed to help potential researchers in exploring areas of research in the field of graphic symbol recognition.

Beyond Homology Transfer: Deep Learning for Automated Annotation of Proteins

Accurate annotation of protein functions is important for a profound understanding of molecular biology. A large number of proteins remain uncharacterized because of the sparsity of available supporting information. For a large set of uncharacterized proteins, the only type of information available is their amino acid sequence. This motivates the need to make sequence based computational techniques that can precisely annotate uncharacterized proteins. In this paper, we propose DeepSeq – a deep learning architecture – that utilizes only the protein sequence information to predict its associated functions. The prediction process does not require handcrafted features; rather, the architecture automatically extracts representations from the input sequence data. Results of our experiments with DeepSeq indicate significant improvements in terms of prediction accuracy when compared with other sequence-based methods. Our deep learning model achieves an overall validation accuracy of 86.72%, with an F1 score of 71.13%. We achieved improved results for protein function prediction problem through DeepSeq, by utilizing sequence only information. Moreover, using the automatically learned features and without any changes to DeepSeq, we successfully solved a different problem i.e. protein function localization, with no human intervention. Finally, we discuss how the same architecture can be used to solve even more complicated problems such as prediction of 2D and 3D structure as well as protein-protein interactions.