Search

We describe a new method based on principal component analysis and robust consensus ensemble clustering to identify and elucidate the subtypes of breast cancer disease. The method was applied to microarray gene expression data using micro-dissection of samples from 36 breast cancer patients with at least two of three pathological stages of disease. Controls were normal breast epithelial cells from 3 disease free patients. Our method identified an optimum set of genes and strong, stable clusters which correlated well with clinical classification into Luminal, Basal and Her2+ subtypes based on ER, PR and Her2 status. It also revealed a hierarchical portrait of disease progression through various grades and stages and identified genes and functional pathways for each stage, grade and disease subtype. We found that gene expression heterogeneity across subtypes is much greater than the heterogeneity of progression from DCIS to IDC within a subtype, suggesting that the disease subtypes are distinct disease processes. The averaging over data perturbations and clustering methods is critical in the robust identification of subtypes and gene markers for grade and progression.

We introduce a new and potentially valuable method for delineating the repertoire of protein complexes in highly mutable organisms and, in conjunction with other methods, for specifying the structural details of complexes. In the first instance the method provides a guide to selecting proteins for co-crystallization; in the second it augments the collection of structures determined by crystallography and other methods, including the discovery of possible alternative binding sites of known complexes. The key to the method is the availability of multiple sequence variants of an organism-arrived at either naturally or by directed mutagenesis in appropriate laboratory facilities. Amino acids that are important for the structural stability of a protein or complex tend to be conserved, generally mutating only when compensatory changes occur. Consequently significant correlations in variation of two conserved amino acids in the same protein suggest that they interact with one another, either directly or indirectly. Similarly, correlated mutations between conserved amino acids in different proteins suggest that they may be at a site of physical interaction. We have identified all highly conserved 9-11 amino acid long segments from HIV proteins and then identified pairs from different proteins with highly significant co-variation. Using the HIV reverse transcriptase and integrase proteins as an example, we demonstrate how the interface and combining sites can be inferred by co-variation analysis and rigid body docking. The potential significance for antiviral drug and vaccine design is briefly discussed.

Phylogenetic profiling is now an effective computational method to detect functional associations between proteins. The method links two proteins in accordance with the similarity of their phyletic distributions across a set of genomes. While pair-wise linkage is useful, it misses correlations in higher order groups: triplets, quadruplets, and so on. Here we assess the probability of observing co-occurrence patterns of 3 binary profiles by chance and show that this probability is asymptotically the same as the mutual information in three profiles. We demonstrate the utility of the probability and the mutual information metrics in detecting overly represented triplets of orthologous proteins which could not be detected using pairwise profiles. These triplets serve as small building blocks, i.e. motifs in protein networks; they allow us to infer the function of uncharacterized members, and facilitate analysis of the local structure and global organization of the protein network. Our method is extendable to N-component clusters, and therefore serves as a general tool for high order protein function annotation.

The common consideration in approaching protection against HIV, whether by a vaccine or therapeutic, is identification of suitable targets. Among the central criteria for suitability is target stability; i.e. resistance to mutation. In this paper we address the problem of stability, and develop methods for identifying stable targets. The targets that we focus on are structures formed by viral peptides and products of the class I major histocompatibility complex, the target of the immune system. The method mines the large databases of fully sequenced HIV genomes and MHC binding peptides, and takes account of human polymorphism to construct hundreds of subpopulation specific stable targets, each consisting of combinations of 3-5 complexes.

Transcription factor binding sites (TFBS) in gene promoter regions are often predicted by using position specific scoring matrices (PSSMs), which summarize sequence patterns of experimentally determined TF binding sites. Although PSSMs are more reliable than simple consensus string matching in predicting a true binding site, they generally result in high numbers of false positive hits. This study attempts to reduce the number of false positive matches and generate new predictions by integrating various types of genomic data by two methods: a Bayesian allocation procedure, and support vector machine classification. Several methods will be explored to strengthen the prediction of a true TFBS in the Saccharomyces cerevisiae genome: binding site degeneracy, binding site conservation, phylogenetic profiling, TF binding site clustering, gene expression profiles, GO functional annotation, and k-mer counts in promoter regions. Binding site degeneracy (or redundancy) refers to the number of times a particular transcription factor's binding motif is discovered in the upstream region of a gene. Phylogenetic conservation takes into account the number of orthologous upstream regions in other genomes that contain a particular binding site. Phylogenetic profiling refers to the presence or absence of a gene across a large set of genomes. Binding site clusters are statistically significant clusters of TF binding sites detected by the algorithm ClusterBuster. Gene expression takes into account the idea that when the gene expression profiles of a transcription factor and a potential target gene are correlated, then it is more likely that the gene is a genuine target. Also, genes with highly correlated expression profiles are often regulated by the same TF(s). The GO annotation data takes advantage of the idea that common transcription targets often have related function. Finally, the distribution of the counts of all k-mers of length 4, 5, and 6 in gene's promoter region were examined as means to predict TF binding. In each case the data are compared to known true positives taken from ChIP-chip data, Transfac, and the Saccharomyces Genome Database. First, degeneracy, conservation, expression, and binding site clusters were examined independently and in combination via Bayesian allocation. Then, binding sites were predicted with a support vector machine (SVM) using all methods alone and in combination. The SVM works best when all genomic data are combined, but can also identify which methods contribute the most to accurate classification. On average, a support vector machine can classify binding sites with high sensitivity and an accuracy of almost 80%.

High-throughput gene expression profiling can identify sets of genes that are differentially expressed between different phenotypes. Discovering marker genes is particularly important in diagnosis of a cancer phenotype. However, gene sets produced to date are too large to be economically viable diagnostics. We use a hybrid decision tree-discriminant analysis to identify small sets of genes, i.e. single genes and gene pairs, which separate normal samples from different stages of tumor samples. Half the samples are selected for training to form the probability distribution of expression values of each gene. The distributions for the tumor and normal phenotypes are then used to classify the test samples. The algorithm also identifies gene pairs by combining the probability distributions to construct a decision tree which is used to determine the class of test samples. After a series of training and testing sessions, genes and gene pairs that classify all samples correctly are recorded. The method was applied to a breast cancer data; and classifier genes that distinguish normal breast from different stages of breast tumor were identified. The genes were ranked according to their minimum Euclidean distance between the expression values in tumor and normal samples. The algorithm was able to pick known cancer related genes but also find genes that were not identified as differentially expressed by t-test with a 2 fold cut-off. Overall, the method generates possible diagnostic genes and gene pairs for a specific disease phenotype to pursue further biological interpretations in cancer biology.

Genome scale identification of transcription factor binding sites (TFBS) is fundamental to understanding the complexities of mRNA expression at both the cell and organismal levels. While high-throughput experimental methods provide associations between transcription factors and the genes they regulate under a specified experimental condition, computational methods are still required to pinpoint the exact location of binding. Moreover, since the binding site is an intrinsic property of the promoter region, computational methods are in principle more general than condition dependent experimental methods. Computational identification of TFBSs is complicated in at least two different ways. First, transcription factors bind a heterogeneous distribution of sites and therefore have a distribution of affinities. Second, the set of sequences for which a common site is to be determined do not all have a site for the TF of interest. In this paper, we evaluate the robustness of TFBS identification with respect to both effects. We show addition of upstream regions that do not have the TFBS destroy the specificity of the predicted binding site. We also propose a method to calculate the distance between position weight matrices that can be used to measure "drift'' from the canonical binding site. The results presented here could be useful in developing future transcription factor binding site identification algorithms.

Problems of inference in systems biology are ideally reduced to formulations which can efficiently represent the features of interest. In the case of predicting gene regulation and pathway networks, an important feature which describes connected genes and proteins is the relationship between active and inactive forms, i.e. between the "on" and "off" states of the components. While not optimal at the limits of resolution, these logical relationships between discrete states can often yield good approximations of the behavior in larger complex systems, where exact representation of measurement relationships may be intractable. We explore techniques for extracting binary state variables from measurement of gene expression, and go on to describe robust measures for statistical significance and information that can be applied to many such types of data. We show how statistical strength and information are equivalent criteria in limiting cases, and demonstrate the application of these measures to simple systems of gene regulation.

ELISA (http://romi.bu.edu/elisa/) is a database that was designed for flexibility in defining interesting queries about protein domain evolution. We have defined and included both the inherent characteristics of the domains such as structure and function and comparisons of these characteristics between domains. Thus, the database is useful in defining structural and functional links between related protein domains and by extension sequences that encode them. In this database we introduce and employ a novel method of functional annotation and comparison. For each protein domain we create a probabilistic functional annotation tree using GO. We have designed an algorithm that accurately compares these trees and thus provides a measure of "functional distance" between two protein domains. Along with functional annotation, we have also included structural comparison between protein domains and best sequence comparisons to all known genomes. The latter enables researchers to dynamically do searches for domains sharing similar phylogenetic profiles. This combination of data and tools enables the researcher to design complex queries to carry out research in the areas of protein domain evolution, structure prediction and functional annotation of novel sequences.

A question of fundamental importance is the definition and identification of modules from microarray experiments. A wide variety of techniques have been used to gain insight into the elucidation of such modules. One problem, however, is the inability to directly compare results between the different data sets produced due to the inherent parameterizations of their approaches. We first aim to provide a mechanism by which different approaches to module finding can be directly compared. Moreover, the same approach can be used to internally compare the modules predicted by the same technique, but at different parameterizations. We apply this approach to analyze the flow of genes through modules at different module thresholds of the Barkai Signature method, thereby further resolving the modules into sets of co-expressed genes.