Computational systems bioinformatics. Computational Systems Bioinformatics Conference最新文献

The systematic inference of biologically relevant influence networks remains a challenging problem in computational biology. Even though the availability of high-throughput data has enabled the use of probabilistic models to infer the plausible structure of such networks, their true interpretation of the biology of the process is questionable. In this work, we propose a network inference methodology, based on the directed information (DTI) criterion, which incorporates the biology of transcription within the framework, so as to enable experimentally verifiable inference. We use publicly available embryonic kidney and T-cell microarray datasets to demonstrate our results. We present two variants of network inference via DTI (supervised and unsupervised) and the inferred networks relevant to mammalian nephrogenesis as well as T-cell activation. We demonstrate the conformity of the obtained interactions with literature as well as comparison with the coefficient of determination (CoD) method. Apart from network inference, the proposed framework enables the exploration of specific interactions, not just those revealed by data.

Motivation: The deconvolution of the relationships between BAC clones and genes is a crucial step in the selective sequencing of the regions of interest in a genome. It usually requires combinatorial pooling of unique probes obtained from the genes (unigenes), and the screening of the BAC library using the pools in a hybridization experiment. Since several probes can hybridize to the same BAC, in order for the deconvolution to be achievable the pooling design has to be able to handle a large number of positives. As a consequence, smaller pools need to be designed which in turn increases the number of hybridization experiments possibly making the entire protocol unfeasible.

Results: We propose a new algorithm that is capable of producing high accuracy deconvolution even in the presence of a weak pooling design, i.e., when pools are rather large. The algorithm compensates for the decrease of information in the hybridization data by taking advantage of a physical map of the BAC clones. We show that the right combination of combinatorial pooling and our algorithm not only dramatically reduces the number of pools required, but also successfully deconvolutes the BAC-gene relationships with almost perfect accuracy.

In recent years, sequence database searching has been conducted through local alignment heuristics, pattern-matching, and comparison of short statistically significant patterns. While these approaches have unlocked many clues as to sequence relationships, they are limited in that they do not provide context-sensitive searching capabilities (e.g. considering pseudoknots, protein binding positions, and complementary base pairs). Stochastic grammars (hidden Markov models HMMs and stochastic context-free grammars SCFG) do allow for flexibility in terms of local context, but the context comes at the cost of increased computational complexity. In this paper we introduce a new grammar based method for searching for RNA motifs that exist within a conserved RNA structure. Our method constrains computational complexity by using a chain of topology elements. Through the use of a case study we present the algorithmic approach and benchmark our approach against traditional methods.

The effectiveness of comparative modeling approaches for protein structure prediction can be substantially improved by incorporating predicted structural information in the initial sequence-structure alignment. Motivated by the approaches used to align protein structures, this paper focuses on developing machine learning approaches for estimating the RMSD value of a pair of protein fragments. These estimated fragment-level RMSD values can be used to construct the alignment, assess the quality of an alignment, and identify high-quality alignment segments. We present algorithms to solve this fragment-level RMSD prediction problem using a supervised learning framework based on support vector regression and classification that incorporates protein profiles, predicted secondary structure, effective information encoding schemes, and novel second-order pairwise exponential kernel functions. Our comprehensive empirical study shows superior results compared to the profile-to-profile scoring schemes.