Georgina Mirceva

Comparative Analysis of Three Efficient Approaches for Retrieving Protein 3D Structures

In this paper, comparative analysis is presented of our three 3D structure-based approaches for the efficient retrieval of protein molecules. All approaches rely on the 3D structure of the proteins. In the first approach, the Spherical Trace Transform is applied to protein 3D structures in order to produce geometry based descriptors. Additionally, some biological properties of the protein are taken, thus forming better integrated descriptor. In the second approach, some modification of the ray based descriptor is applied on the backbone of the protein molecule. In the third approach, wavelet transformation is applied on the distance matrix of the Calpha atoms which form the backbone of the protein. The SCOP database was used to evaluate the retrieval accuracy. We provide some experimental results of the retrieval accuracy of our three approaches. The results show that the ray based approach gives the best retrieval accuracy (97,5%), while it is simpler and faster than the other two approaches.

Protein Classification by Matching 3D Structures

In this paper, a 3D structure-based approach is presented for the efficient classification of protein molecules. The method relies on the geometric 3D structure of the proteins. After proper positioning of the 3D structures, the spherical trace transform is applied to them to produce geometry - based descriptors, which are completely rotation invariant. Additionally, some biological properties of the protein are taken, and added to the geometry-based descriptor, thus forming better integrated descriptor. We have used nearest neighbour classification on the previously extracted descriptors. A part of the FSSP/DALI database, was used to evaluate the classification accuracy of this system. The results show that this method achieves more than 92 percent classification accuracy while it is simpler and faster than the DALI method. We provide some experimental results of the implemented system.

Efficient Approaches for Retrieving Protein Tertiary Structures

The 3D conformation of a protein in the space is the main factor which determines its function in living organisms. Due to the huge amount of newly discovered proteins, there is a need for fast and accurate computational methods for retrieving protein structures. Their purpose is to speed up the process of understanding the structure-to-function relationship which is crucial in the development of new drugs. There are many algorithms addressing the problem of protein structure retrieval. In this paper, we present several novel approaches for retrieving protein tertiary structures. We present our voxel-based descriptor. Then we present our protein ray-based descriptors which are applied on the interpolated protein backbone. We introduce five novel wavelet descriptors which perform wavelet transforms on the protein distance matrix. We also propose an efficient algorithm for distance matrix alignment named Matrix Alignment by Sequence Alignment within Sliding Window (MASASW), which has shown as much faster than DALI, CE, and MatAlign. We compared our approaches between themselves and with several existing algorithms, and they generally prove to be fast and accurate. MASASW achieves the highest accuracy. The ray and wavelet-based descriptors as well as MASASW are more accurate than CE.

A Novel Fuzzy Decision Tree Based Method for Detecting Protein Active Sites

The knowledge of the functions of protein structures is essential for development of new drugs, better crops and synthetic biochemical. There are numerous experimental methods for determining the protein functions, but because of their complexity the number of protein molecules with undetermined functions is rapidly growing. Thus, there is an evident need for development of computer methods for determining the functions of the protein structures. In this study, we introduce the fuzzy theory for protein active sites detection. We propose a novel fuzzy decision tree (FDT) based method for predicting protein active sites that later could be used for determining the functions of the protein molecules. First, we extract several characteristics of the amino acids. Then, we induce FDTs that would be used to predict the protein active sites. We provide experimental results of the evaluation of the prediction power of the proposed method. Also, our method is compared with other machine learning techniques that could be used for this purpose.

Top-Down Approach for Protein Binding Sites Prediction Based on Fuzzy Pattern Trees

The understanding of the relation between the protein structure and protein functions is one of the main research topics in bioinformatics nowadays. Due to the complexity of the methods for determining protein functions, there are many proteins with unknown functions. Hence, many researchers investigate various computational methods for determining protein functions. We focus on investigating methods for predicting the protein binding sites, and afterwards their characteristics could be used for annotating protein structures. In order to overcome the problem of sensitivity on data changes, we already introduced the fuzzy theory for protein biding sites prediction. In this paper we introduce an approach for detecting protein binding sites using a top-down induction of fuzzy pattern trees. This approach outperforms the existing bottom-up approach for inducing fuzzy pattern trees, and also most of the examined approaches which are based on classical classification algorithms.

Method for Protein Active Sites Detection Based on Fuzzy Decision Trees

The knowledge of the protein functions is very important in the development of new drugs. Many experimental methods for determining protein function exist, but due to their complexity the number of protein structures with unknown functions is rapidly growing. So, there is an obvious necessity for development of computer methods for annotating protein structures. In this paper we present a fuzzy decision tree based method for protein active sites detection, which could be used for annotating protein structures. We extract several features of the amino acids, and then using different membership functions we build fuzzy decision trees in order to detect the possible active sites. We provide some experimental results of the evaluation of our method. Additionally, our method is compared with several existing methods for protein active sites detection.

Incorporating several features in the protein ray descriptor for more accurate protein 3D structure retrieval

The retrieval of protein structures is one of the most popular topics in bioinformatics community nowadays, since it can be further used for determining protein function. With the technology innovation, the number of protein tertiary structures increases every day, so the necessity of accurate, robust and efficient algorithms for protein 3D structure retrieval arises. In this paper we improve the ray based descriptor, in sense of achieving higher precision. First, the protein backbone is uniformly interpolated, as in the process of extraction of the existing protein ray based descriptor. Then, for each of the approximation points, beside the Euclidean distance from the point to the centre of mass, the ASA, RASA and hydrophobicity of the nearest amino acid are extracted and added to the descriptor. The analysis showed that by incorporating these features, we achieve higher precision (2.26% improvement) than by using only the Euclidean distance as a single feature.

Protein Classification Based on 3D Structures and Fractal Features

To understand the structure-to-function relationship, life sciences researchers and biologists need to retrieve similar structures and classify them into the same protein fold. In this paper, we propose a 3D structure-based approach for efficient classification of protein molecules. Classification is performed in three phases. In the first phase, we apply fractal descriptor matching as a filter. Then, protein structures which satisfy the fractal and radius tolerance are classified in the second phase. In this phase, 3D Fourier Transform is applied in order to produce rotation invariant descriptors. Additionally, some properties of primary and secondary structure are taken. In the third phase we use k nearest neighbor classifier. Our approach achieves 86% classification accuracy with applying fractal filter, and 92% without fractal filter. It is shown that fractal filter significantly shorten the classification time. Our system is faster (seconds) than DALI system (minutes, hours, days), and we still get satisfactory results.

Influence of Algebraic T-norm on Different Indiscernibility Relationships in Fuzzy-Rough Rule Induction Algorithms

The rule induction algorithms generate rules directly in human-understandable if-then form, and this property is essential of successful intelligent classifier. Similar as crisp algorithms, the fuzzy and rough set methods are used to generate rule based induction algorithms. Recently, a rule induction algorithms based on fuzzy-rough theory were proposed. These algorithms operate on the well-known upper and lower approximation concepts, and they are sensitive to different T-norms, implicators and more over; to different similarity metrics. In this paper, we experimentally evaluate the influence of the T-norm Algebraic norm on the classification and regression tasks performance on three fuzzy-rough rule induction algorithms. The experimental results revealed some interesting results, moreover, the choice of similarity metric in combination with the T-norm on some datasets has no influence at all. Based on the experimental results, further investigation is required to investigate the influence of other T-norms on the algorithm’s performance.

Influence of Fuzzy Tolerance Metrics on Classification and Regression Tasks for Fuzzy-Rough Nearest Neighbour Algorithms

In this paper, we investigate the influence of the fuzzy tolerance relationship (fuzzy similarity metrics) on two fuzzy and two fuzzy-rough nearest neighbour algorithms for both classification and regression tasks. The fuzzy similarity metric plays a major role in construction of the lower and upper approximations of decision classes, and therefore has high influence on the accuracy of the algorithm. The experimental results evaluated on the four approaches show the difficulty to estimate a single metric that will be good in all cases. Moreover, the choice of similarity metric on some datasets has not influence at all. This require further investigation, not only with similarity metrics, but also for evaluating the algorithms with different T-norms and implicators.