Cem Meydan

Prediction of peptides binding to MHC class I and II alleles by temporal motif mining

BackgroundMHC (Major Histocompatibility Complex) is a key player in the immune response of most vertebrates. The computational prediction of whether a given antigenic peptide will bind to a specific MHC allele is important in the development of vaccines for emerging pathogens, the creation of possibilities for controlling immune response, and for the applications of immunotherapy. One of the problems that make this computational prediction difficult is the detection of the binding core region in peptides, coupled with the presence of bulges and loops causing variations in the total sequence length. Most machine learning methods require the sequences to be of the same length to successfully discover the binding motifs, ignoring the length variance in both motif mining and prediction steps. In order to overcome this limitation, we propose the use of time-based motif mining methods that work position-independently.ResultsThe prediction method was tested on a benchmark set of 28 different alleles for MHC class I and 27 different alleles for MHC class II. The obtained results are comparable to the state of the art methods for both MHC classes, surpassing the published results for some alleles. The average prediction AUC values are 0.897 for class I, and 0.858 for class II.ConclusionsTemporal motif mining using partial periodic patterns can capture information about the sequences well enough to predict the binding of the peptides and is comparable to state of the art methods in the literature. Unlike neural networks or matrix based predictors, our proposed method does not depend on peptide length and can work with both short and long fragments. This advantage allows better use of the available training data and the prediction of peptides of uncommon lengths.

Testing robustness of relative complexity measure method constructing robust phylogenetic trees for Galanthus L. Using the relative complexity measure

BackgroundMost phylogeny analysis methods based on molecular sequences use multiple alignment where the quality of the alignment, which is dependent on the alignment parameters, determines the accuracy of the resulting trees. Different parameter combinations chosen for the multiple alignment may result in different phylogenies. A new non-alignment based approach, Relative Complexity Measure (RCM), has been introduced to tackle this problem and proven to work in fungi and mitochondrial DNA.ResultIn this work, we present an application of the RCM method to reconstruct robust phylogenetic trees using sequence data for genus Galanthus obtained from different regions in Turkey. Phylogenies have been analyzed using nuclear and chloroplast DNA sequences. Results showed that, the tree obtained from nuclear ribosomal RNA gene sequences was more robust, while the tree obtained from the chloroplast DNA showed a higher degree of variation.ConclusionsPhylogenies generated by Relative Complexity Measure were found to be robust and results of RCM were more reliable than the compared techniques. Particularly, to overcome MSA-based problems, RCM seems to be a reasonable way and a good alternative to MSA-based phylogenetic analysis. We believe our method will become a mainstream phylogeny construction method especially for the highly variable sequence families where the accuracy of the MSA heavily depends on the alignment parameters.

Multiple sequence alignment based on structural properties

A multiple sequence alignment (MSA) is a sequence alignment of three or more biological sequences. Main idea behind multiple sequence alignment is to see the similarities between input sequences, to be able to make phylogenetic analysis and other evolutionary conclusions. We propose a multiple sequence alignment method based on contact maps derived from structural data and network properties. We show that such methods may be useful in creating multiple alignments that can identify domains and similar structures where sequence identity is low.

Biomarker discovery for toxicity

Toxicity biomarkers allow the safe evaluation of possible toxic effects of a substance in early phases of drug discovery. Finding the optimal subset of genes to use as biomarkers is an important problem. We tried evolutionary classification methods for finding biomarkers in hexachlorobenzene (HCB) toxicity using microarray data. We improve upon Kucukural et al. by modifying the algorithm to incrementally filter the features instead of generating new populations from scratch and by finding the common subset of features from multiple runs to be used as biomarkers. Using this modified genetic algorithm, we discovered gene sets of size 4 that were able to predict HCB exposure with >99% accuracy in 5-fold cross-validation tests. Repeating this process on independent test studies resulted in 14 biologically significant genes that predict exposure with 91% accuracy, surpassing other feature selection methods. Making use of these genes as biomarkers may allow us to detect hepatotoxic substances similar to HCB in a fast and cost-efficient manner when there are no emerging symptoms.

Discovery of Biomarkers for Hexachlorobenzene Toxicity Using Population Based Methods on Gene Expression Data

Discovering toxicity biomarkers is important in drug discovery to safely evaluate possible toxic effects of a substance in early phases. We tried evolutionary classification methods for selecting the important classifier genes in hexachlorobenzene toxicity using microarray data. Using modified genetic algorithms for selection of minimum number of features for classification of gene expression data, we discovered a number of gene sets of size 4 that were able to discriminate between the control and the hexachlorobenzene (HCB) exposed group of Brown-Norway rats with >99% accuracy in 5-fold cross-validation tests, whereas classification using all of the genes with SVM and other methods yielded results that vary between 48.48% to 81.81%. Making use of this small number of genes as biomarkers may allow us to detect toxicity of substances with mechanisms of toxicity similar to HCB in a fast and cost efficient manner when there are no emerging symptoms.

Representation of protein secondary structure using bond-orientational order parameters

Structural studies of proteins for motif mining and other pattern recognition techniques require the abstraction of the structure into simpler elements for robust matching. In this study, we propose the use of bond-orientational order parameters, a well-established metric usually employed to compare atom packing in crystals and liquids. Creating a vector of orientational order parameters of residue centers in a sliding window fashion provides us with a descriptor of local structure and connectivity around each residue that is easy to calculate and compare. To test whether this representation is feasible and applicable to protein structures, we tried to predict the secondary structure of protein segments from those descriptors, resulting in 0.99 AUC (area under the ROC curve). Clustering those descriptors to 6 clusters also yield 0.93 AUC, showing that these descriptors can be used to capture and distinguish local structural information.

Prediction of Peptides Binding to MHC Class I Alleles by Partial Periodic Pattern Mining

MHC (Major Histocompatibility Complex) is a key player in the immune response of an organism. It is important to be able to predict which antigenic peptides will bind to a specific MHC allele and which will not, creating possibilities for controlling immune response and for the applications of immunotherapy. However, a problem for MHC class I is the presence of bulges and loops in the peptides, changing the total length. Most machine learning methods in use today require the sequences to be of same length to successfully mine the binding motifs. We propose the use of time-based data mining methods in motif mining to be able to mine motifs position-independently. Also, the information for both binding and non-binding peptides is used on the contrary to the other methods which only rely on binding peptides. The prediction results are between 60-95% for the tested alleles.

Optimization of morphological data in numerical taxonomy analysis using genetic algorithms feature selection method

Studies in Numerical Taxonomy are carried out by measuring characters as much as possible. The workload over scientists and labor to perform measurements will increase proportionally with the number of variables (or characters) to be used in the study. However, some part of the data may be irrelevant or sometimes meaningless. Here in this study, we introduce an algorithm to obtain a subset of data with minimum characters that can represent original data. Morphological characters were used in optimization of data by Genetic Algorithms Feature Selection method. The analyses were performed on an 18 character*11 taxa data matrix with standardized continuous characters. The analyses resulted in a minimum set of 2 characters, which means the original tree based on the complete data can also be constructed by those two characters.