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

DMSS (Discrete Metabolic Simulation System) is a framework for modelling and simulating metabolic pathways. Quantitative simulation of metabolic pathways is achieved using discrete-event techniques. The approach differs from most quantitative simulators of metabolism which employ either time-differentiated functions or mathematical modelling techniques. Instead, models are constructed from biochemical data and biological knowledge, with accessibility and relevance to biologists serving as key features of the system.

The statistical significance of gapped local alignments is characterized by analyzing the extremal statistics of the scores obtained from the alignment of random amino acid sequences. By identifying a complete set of linked clusters, "islands," we devise a method which accurately predicts the extremal score statistics by using only one to a few pairwise alignments. The success of our method relies crucially on the link between the statistics of island scores and extremal score statistics. This link is motivated by heuristic arguments, and firmly established by extensive numerical simulations for a variety of scoring parameter settings and sequence lengths. Our approach is several orders of magnitude faster than the widely used shuffling method, since island counting is trivially incorporated into the basic Smith-Waterman alignment algorithm with minimal computational cost, and all islands are counted in a single alignment. The availability of a rapid and accurate significance estimation method gives one the flexibility to fine tune scoring parameters to detect weakly homologous sequences and obtain optimal alignment fidelity.

In molecular databases, structural classification is a basic task that can be successfully approached by nearest neighbor methods. The underlying similarity models consider spatial properties such as shape and extension as well as thematic attributes. We introduce 3D shape histograms as an intuitive and powerful approach to model similarity for solid objects such as molecules. Errors of measurement, sampling, and numerical rounding may result in small displacements of atomic coordinates. These effects may be handled by using quadratic form distance functions. An efficient processing of similarity queries based on quadratic forms is supported by a filter-refinement architecture. Experiments on our 3D protein database demonstrate the high classification accuracy of more than 90% and the good performance of the technique.

Recently, there has been much effort in making databases for molecular biology more accessible and interoperable. However, information in text form, such as MEDLINE records, remains a greatly underutilized source of biological information. We have begun a research effort aimed at automatically mapping information from text sources into structured representations, such as knowledge bases. Our approach to this task is to use machine-learning methods to induce routines for extracting facts from text. We describe two learning methods that we have applied to this task--a statistical text classification method, and a relational learning method--and our initial experiments in learning such information-extraction routines. We also present an approach to decreasing the cost of learning information-extraction routines by learning from "weakly" labeled training data.

In an earlier paper, we described a new method for phylogenetic tree reconstruction called the Disk Covering Method, or DCM. This is a general method which can be used with any existing phylogenetic method in order to improve its performance. We showed analytically and experimentally that when DCM is used in conjunction with polynomial time distance-based methods, it improves the accuracy of the trees reconstructed. In this paper, we discuss a variant on DCM, that we call DCM2. DCM2 is designed to be used with phylogenetic methods whose objective is the solution of NP-hard optimization problems. We show that DCM2 can be used to accelerate searches for Maximum Parsimony trees. We also motivate the need for solutions to NP-hard optimization problems by showing that on some very large and important datasets, the most popular (and presumably best performing) polynomial time distance methods have poor accuracy.

In this paper, we describe our algorithmic approach to constructing an alignment of (contiging) a set of restriction maps created from the images of individual genomic (uncloned) DNA molecules digested by restriction enzymes. Generally, these DNA segments are sized in the range of 1-4 Mb. The goal is to devise contiging algorithms capable of producing high-quality composite maps rapidly and in a scaleable manner. The resulting software is a key component of our physical mapping automation tools and has been used to create complete maps of various microorganisms (E. coli, P. falciparum and D. radiodurans). Experimental results match known sequence data.

In the course of our efforts to build extended regions of human genomic sequence by assembling individual BAC sequences, we have encountered several instances where a region of the genome has been sequenced independently using reagents derived from two different individuals. Comparing these sequences allows us to analyze the frequency and distribution of single nucleotide polymorphisms (SNPs) in the human genome. The observed transition/transversion frequencies are consistent with a biological origin for the sequence discrepancies, and this suggests that the data produced by large sequencing centers are accurate enough to be used as the basis for SNP analysis. The observed distribution of single nucleotide polymorphisms in the human genome is not uniform. An apparent duplication in the human genome extending over more than 130 kb between chromosomes 1p34 and 16p13 is reported. Independently derived sequences covering these regions are more than 99.9% identical, indicating that this duplication event must have occurred quite recently. FISH mapping results reported by the relevant laboratories indicate that the human population may be polymorphic for this duplication. We present a population genetic theory for the expected distribution of SNPs and derive an algorithm for probabilistically segmenting genomic sequence into regions that are identical by descent (IBD) between two individuals based on this theory and the observed locations of polymorphisms. Based on these methods and a random mating model for the human population, estimates are made for the mutation rate in the human genome.

A new method, called the Fisher kernel method, for detecting remote protein homologies is introduced and shown to perform well in classifying protein domains by SCOP superfamily. The method is a variant of support vector machines using a new kernel function. The kernel function is derived from a hidden Markov model. The general approach of combining generative models like HMMs with discriminative methods such as support vector machines may have applications in other areas of biosequence analysis as well.

Simulated data sets have been found to be useful in developing software systems because (1) they allow one to study the effect of a particular phenomenon in isolation, and (2) one has complete information about the true solution against which to measure the results of the software. In developing a software suite for assembling a whole human genome shotgun data set, we have developed a simulator, celsim, that permits one to describe and stochastically generate a target DNA sequence with a variety of repeat structures, to further generate polymorphic variants if desired, and to generate a shotgun data set that might be sampled from the target sequence(s). We have found the tool invaluable and quite powerful, yet the design is extremely simple, employing a special type of stochastic grammar.