Genome informatics. International Conference on Genome Informatics最新文献

Many theoretical advances have been made in applying probabilistic inference methods to improve the power of sequence homology searches, yet the BLAST suite of programs is still the workhorse for most of the field. The main reason for this is practical: BLAST's programs are about 100-fold faster than the fastest competing implementations of probabilistic inference methods. I describe recent work on the HMMER software suite for protein sequence analysis, which implements probabilistic inference using profile hidden Markov models. Our aim in HMMER3 is to achieve BLAST's speed while further improving the power of probabilistic inference based methods. HMMER3 implements a new probabilistic model of local sequence alignment and a new heuristic acceleration algorithm. Combined with efficient vector-parallel implementations on modern processors, these improvements synergize. HMMER3 uses more powerful log-odds likelihood scores (scores summed over alignment uncertainty, rather than scoring a single optimal alignment); it calculates accurate expectation values (E-values) for those scores without simulation using a generalization of Karlin/Altschul theory; it computes posterior distributions over the ensemble of possible alignments and returns posterior probabilities (confidences) in each aligned residue; and it does all this at an overall speed comparable to BLAST. The HMMER project aims to usher in a new generation of more powerful homology search tools based on probabilistic inference methods.

We have systematically analyzed various topological patterns comprising 1, 2 or 3 nodes in the mammalian metabolic, signal transduction and transcription networks: These patterns were analyzed with regard to their frequency and statistical over-representation in each network, as well as to their topological significance for the coherence of the networks. The latter property was evaluated using the pairwise disconnectivity index, which we have recently introduced to quantify how critical network components are for the internal connectedness of a network. The 1-node pattern made up by a vertex with a self-loop has been found to exert particular properties in all three networks. In general, vertices with a self-loop tend to be topologically more important than other vertices. Moreover, self-loops have been found to be attached to most 2-node and 3-node patterns, thereby emphasizing a particular role of self-loop components in the architectural organization of the networks. For none of the networks, a positive correlation between the mean topological significance and the Z-score of a pattern could be observed. That is, in general, motifs are not per se more important for the overall network coherence than patterns that are not over-represented. All 2- and 3-node patterns that are over-represented and thus qualified as motifs in all three networks exhibit a loop structure. This intriguing observation can be viewed as an advantage of loop-like structures in building up the regulatory circuits of the whole cell. The transcription network has been found to differ from the other networks in that (i) self-loops play an even higher role, (ii) its binary loops are highly enriched with self-loops attached, and (iii) feed-back loops are not over-represented. Metabolic networks reveal some particular topological properties which may reflect the fact that metabolic paths are, to a large extent, reversible. Interestingly, some of the most important 3-node patterns of both the transcription and the signaling network can be concatenated to subnetworks comprising many genes that play a particular role in the regulation of cell proliferation.

Local sequence-structure relationships in the loop regions of proteins were comprehensively estimated using simple prediction tools based on support vector regression (SVR). End-to-end distance was selected as a rough structural property of fragments, and the end-to-end distances of an enormous number of loop fragments from a wide variety of protein folds were directly predicted from sequence information by using SVR. We found that our method was more accurate than random prediction for predicting the structure of fragments comprising 5, 9, and 17 amino acids; moreover, the extended loop fragments could be successfully distinguished from turn structures on the basis of their sequences, which implies that the sequence-structure relationships were significant for loop fragments with a wide range of end-to-end distances. These results suggest that many loop regions as well as helices and strands restrict the conformational space of the entire tertiary structure of proteins to some extent; moreover, our findings throw light on the mechanism of protein folding and prediction of the tertiary structure of proteins without using structural templates.

Next generation sequencing technologies enable rapid, large-scale production of sequence data sets. Unfortunately these technologies also have a non-neglible sequencing error rate, which biases their outputs by introducing false reads and reducing the quantity of the real reads. Although methods developed for SAGE data can reduce these false counts to a considerable degree, until now they have not been implemented in a scalable way. Recently, a program named FREC has been developed to address this problem for next generation sequencing data. In this paper, we introduce RECOUNT, our implementation of an Expectation Maximization algorithm for tag count correction and compare it to FREC. Using both the reference genome and simulated data, we find that RECOUNT performs as well or better than FREC, while using much less memory (e.g. 5GB vs. 75GB). Furthermore, we report the first analysis of tag count correction with real data in the context of gene expression analysis. Our results show that tag count correction not only increases the number of mappable tags, but can make a real difference in the biological interpretation of next generation sequencing data. RECOUNT is an open-source C++ program available at http://seq.cbrc.jp/recount.

This paper investigates applying statistical topic models to extract and predict relationships between biological entities, especially protein mentions. A statistical topic model, Latent Dirichlet Allocation (LDA) is promising; however, it has not been investigated for such a task. In this paper, we apply the state-of-the-art Collapsed Variational Bayesian Inference and Gibbs Sampling inference to estimating the LDA model. We also apply probabilistic Latent Semantic Analysis (pLSA) as a baseline for comparison, and compare them from the viewpoints of log-likelihood, classification accuracy and retrieval effectiveness. We demonstrate through experiments that the Collapsed Variational LDA gives better results than the others, especially in terms of classification accuracy and retrieval effectiveness in the task of the protein-protein relationship prediction.