Proceedings. International Conference on Intelligent Systems for Molecular Biology最新文献

Several microarray technologies that monitor the level of expression of a large number of genes have recently emerged. Given DNA-microarray data for a set of cells characterized by a given phenotype and for a set of control cells, an important problem is to identify "patterns" of gene expression that can be used to predict cell phenotype. The potential number of such patterns is exponential in the number of genes. In this paper, we propose a solution to this problem based on a supervised learning algorithm, which differs substantially from previous schemes. It couples a complex, non-linear similarity metric, which maximizes the probability of discovering discriminative gene expression patterns, and a pattern discovery algorithm called SPLASH. The latter discovers efficiently and deterministically all statistically significant gene expression patterns in the phenotype set. Statistical significance is evaluated based on the probability of a pattern to occur by chance in the control set. Finally, a greedy set covering algorithm is used to select an optimal subset of statistically significant patterns, which form the basis for a standard likelihood ratio classification scheme. We analyze data from 60 human cancer cell lines using this method, and compare our results with those of other supervised learning schemes. Different phenotypes are studied. These include cancer morphologies (such as melanoma), molecular targets (such as mutations in the p53 gene), and therapeutic targets related to the sensitivity to an anticancer compounds. We also analyze a synthetic data set that shows that this technique is especially well suited for the analysis of sub-phenotype mixtures. For complex phenotypes, such as p53, our method produces an encouragingly low rate of false positives and false negatives and seems to outperform the others. Similar low rates are reported when predicting the efficacy of experimental anticancer compounds. This counts among the first reported studies where drug efficacy has been successfully predicted from large-scale expression data analysis.

In many of the chemical reactions in living cells, enzymes act as catalysts in the conversion of certain compounds (substrates) into other compounds (products). Comparative analyses of the metabolic pathways formed by such reactions give important information on their evolution and on pharmacological targets (Dandekar et al. 1999). Each of the enzymes that constitute a pathway is classified according to the EC (Enzyme Commission) numbering system, which consists of four sets of numbers that categorize the type of the chemical reaction catalyzed. In this study, we consider that reaction similarities can be expressed by the similarities between EC numbers of the respective enzymes. Therefore, in order to find a common pattern among pathways, it is desirable to be able to use the functional hierarchy of EC numbers to express the reaction similarities. In this paper, we propose a multiple alignment algorithm utilizing information content that is extended to symbols having a hierarchical structure. The effectiveness of our method is demonstrated by applying the method to pathway analyses of sugar, DNA and amino acid metabolisms.

Most computational models of protein-ligand interactions consider only the energetics of the final bound state of the complex and do not examine the dynamics of the ligand as it enters the binding site. We have developed a novel technique for studying the dynamics of protein-ligand interactions based on motion planning algorithms from the field of robotics. Our algorithm uses electrostatic and van der Waals potentials to compute the most energetically favorable path between any given initial and goal ligand configurations. We use probabilistic motion planning to sample the distribution of possible paths to a given goal configuration and compute an energy-based "difficulty weight" for each path. By statistically averaging this weight over several randomly generated starting configurations, we compute the relative difficulty of entering and leaving a given binding configuration. This approach yields details of the energy contours around the binding site and can be used to characterize and predict good binding sites. Results from tests with three protein-ligand complexes indicate that our algorithm is able to detect energy barriers around the true binding site that distinguish this site from other predicted low-energy binding sites.

We describe the basic design of a system for automatic detection of protein-protein interactions extracted from scientific abstracts. By restricting the problem domain and imposing a number of strong assumptions which include pre-specified protein names and a limited set of verbs that represent actions, we show that it is possible to perform accurate information extraction. The performance of the system is evaluated with different cases of real-world interaction networks, including the Drosophila cell cycle control. The results obtained computationally are in good agreement with current biological knowledge and demonstrate the feasibility of developing a fully automated system able to describe networks of protein interactions with sufficient accuracy.

Given a sequence of real numbers ("scores"), we present a practical linear time algorithm to find those nonoverlapping, contiguous subsequences having greatest total scores. This improves on the best previously known algorithm, which requires quadratic time in the worst case. The problem arises in biological sequence analysis, where the high-scoring subsequences correspond to regions of unusual composition in a nucleic acid or protein sequence. For instance, Altschul, Karlin, and others have used this approach to identify transmembrane regions, DNA binding domains, and regions of high charge in proteins.

One of the problems associated with the large-scale analysis of unannotated, low quality EST sequences is the detection of coding regions and the correction of frameshift errors that they often contain. We introduce a new type of hidden Markov model that explicitly deals with the possibility of errors in the sequence to analyze, and incorporates a method for correcting these errors. This model was implemented in an efficient and robust program, ESTScan. We show that ESTScan can detect and extract coding regions from low-quality sequences with high selectivity and sensitivity, and is able to accurately correct frameshift errors. In the framework of genome sequencing projects, ESTScan could become a very useful tool for gene discovery, for quality control, and for the assembly of contigs representing the coding regions of genes.

Feed forward neural networks are compared with standard and new statistical classification procedures for the classification of proteins. We applied logistic regression, an additive model and projection pursuit regression from the methods based on a posterior probabilities; linear, quadratic and a flexible discriminant analysis from the methods based on class conditional probabilities, and the K-nearest-neighbors classification rule. Both, the apparent error rate obtained with the training sample (n = 143) and the test error rate obtained with the test sample (n = 125) and the 10-fold cross validation error were calculated. We conclude that some of the standard statistical methods are potent competitors to the more flexible tools of machine learning.

Crystallographic studies play a major role in current efforts towards protein structure determination. Despite recent advances in computational tools for molecular modeling and graphics, the construction of a three-dimensional protein backbone model from crystallographic data remains complex and time-consuming. This paper describes a unique contribution to an automated approach to protein model construction and evaluation, where a model is represented as an annotated trace (or partial trace) of a structure. Candidate models are derived through a topological analysis of the electron density map of a protein. Using sequence alignment techniques, we determine an optimal threading of the known sequence onto the candidate protein structure models. In this threading, connected nodes on the model are associated with adjacent amino acids in the sequence and a fitness score is assigned based on features extracted from the electron density map for the protein. Experimental results demonstrate that crystallographic threading provides an effective means for evaluating the "goodness" of experimentally derived protein models.

We present an analysis of multi-aligned eukaryotic and procaryotic small subunit rRNA sequences using a novel segmentation and clustering procedure capable of extracting subsets of sequences that share common sequence features. This procedure consists of: i) segmentation of aligned sequences using a dynamic programming procedure, and subsequent identification of likely conserved segments; ii) for each putative conserved segment, extraction of a locall homogeneous cluster using a novel polynomial procedure; and iii) intersection of clusters associated with each conserved segment. Aside from their utilit in processing large gap-filled multi-alignments, these algorithms can be applied to a broad spectrum of rRNA analysis functions such as subalignment, phylogenetic subtree extraction and construction, and organism tree-placement, and can serve as a framework to organize sequence data in an efficient and easily searchable manner. The sequence classification we obtained using the method presented here shows a remarkable consistency with the independently constructed eukaryotic phylogenetic tree.