Knowledge-based analysis of microarray gene expression data by using support vector machines

We introduce a method of functionally classifying genes by using gene expression data from DNA microarray hybridization experiments. The method is based on the theory of support vector machines (SVMs). SVMs are considered a supervised computer learning method because they exploit prior knowledge of gene function to identify unknown genes of similar function from expression data. SVMs avoid several problems associated with unsupervised clustering methods, such as hierarchical clustering and self-organizing maps. SVMs have many mathematical features that make them attractive for gene expression analysis, including their flexibility in choosing a similarity function, sparseness of solution when dealing with large data sets, the ability to handle large feature spaces, and the ability to identify outliers. We test several SVMs that use different similarity metrics, as well as some other supervised learning methods, and find that the SVMs best identify sets of genes with a common function using expression data. Finally, we use SVMs to predict functional roles for uncharacterized yeast ORFs based on their expression data.

Computer-based analysis of the binding steps in protein complex formation

Computer models were used to examine whether and under what conditions the multimeric protein complex is inhibited by high concentrations of one of its components—an effect analogous to the prozone phenomenon in precipitin tests. A series of idealized simple “ball-and-stick” structures representing small oligomeric complexes of protein molecules formed by reversible binding reactions were analyzed to determine the binding steps leading to each structure. The equilibrium state of each system was then determined over a range of starting concentrations and Kds and the steady-state concentration of structurally complete oligomer calculated for each situation. A strong inhibitory effect at high concentrations was shown by any protein molecule forming a bridge between two or more separable parts of the complex. By contrast, proteins linked to the outside of the complex by a single bond showed no inhibition whatsoever at any concentration. Nonbridging, multivalent proteins in the body of the complex could show an inhibitory effect or not depending on the structure of the complex and the strength of its bonds. On the basis of this study, we suggest that the prozone phenomenon will occur widely in living cells and that it could be a crucial factor in the regulation of protein complex formation.

Natural genetic exchange between Haemophilus and Neisseria: Intergeneric transfer of chromosomal genes between major human pathogens

Members of the bacterial families Haemophilus and Neisseria, important human pathogens that commonly colonize the nasopharynx, are naturally competent for DNA uptake from their environment. In each genus this process is discriminant in favor of its own and against foreign DNA through sequence specificity of DNA receptors. The Haemophilus DNA uptake apparatus binds a 29-bp oligonucleotide domain containing a highly conserved 9-bp core sequence, whereas the neisserial apparatus binds a 10-bp motif. Each motif (“uptake sequence”, US) is highly over-represented in the chromosome of the corresponding genus, particularly concentrated with core sequences in inverted pairs forming gene terminators. Two Haemophilus core USs were unexpectedly found forming the terminator of sodC in Neisseria meningitidis (meningococcus), and sequence analysis strongly suggests that this virulence gene, located next to IS1106, arose through horizontal transfer from Haemophilus. By using USs as search strings in a computer-based analysis of genome sequence, it was established that while USs of the “wrong” genus do not occur commonly in Neisseria or Haemophilus, where they do they are highly likely to flag domains of chromosomal DNA that have been transferred from Haemophilus. Three independent domains of Haemophilus-like DNA were found in the meningococcal chromosome, associated respectively with the virulence gene sodC, the bio gene cluster, and an unidentified orf. This report identifies intergenerically transferred DNA and its source in bacteria, and further identifies transformation with heterologous chromosomal DNA as a way of establishing potentially important chromosomal mosaicism in these pathogenic bacteria.

A genomic approach to gene fusion technology

Gene expression profiling provides powerful analyses of transcriptional responses to cellular perturbation. In contrast to DNA array-based methods, reporter gene technology has been underused for this application. Here we describe a genomewide, genome-registered collection of Escherichia coli bioluminescent reporter gene fusions. DNA sequences from plasmid-borne, random fusions of E. coli chromosomal DNA to a Photorhabdus luminescens luxCDABE reporter allowed precise mapping of each fusion. The utility of this collection covering about 30% of the transcriptional units was tested by analyzing individual fusions representative of heat shock, SOS, OxyR, SoxRS, and cya/crp stress-responsive regulons. Each fusion strain responded as anticipated to environmental conditions known to activate the corresponding regulatory circuit. Thus, the collection mirrors E. coli's transcriptional wiring diagram. This genomewide collection of gene fusions provides an independent test of results from other gene expression analyses. Accordingly, a DNA microarray-based analysis of mitomycin C-treated E. coli indicated elevated expression of expected and unanticipated genes. Selected luxCDABE fusions corresponding to these up-regulated genes were used to confirm or contradict the DNA microarray results. The power of partnering gene fusion and DNA microarray technology to discover promoters and define operons was demonstrated when data from both suggested that a cluster of 20 genes encoding production of type I extracellular polysaccharide in E. coli form a single operon.

Urokinase-type plasminogen activator is effective in fibrin clearance in the absence of its receptor or tissue-type plasminogen activator.

The availability of gene-targeted mice deficient in the urokinase-type plasminogen activator (uPA), urokinase receptor (uPAR), tissue-type plasminogen activator (tPA), and plasminogen permits a critical, genetic-based analysis of the physiological and pathological roles of the two mammalian plasminogen activators. We report a comparative study of animals with individual and combined deficits in uPAR and tPA and show that these proteins are complementary fibrinolytic factors in mice. Sinusoidal fibrin deposits are found within the livers of nearly all adult mice examined with a dual deficiency in uPAR and tPA, whereas fibrin deposits are never found in livers collected from animals lacking uPAR and rarely detected in animals lacking tPA alone. This is the first demonstration that uPAR has a physiological role in fibrinolysis. However, uPAR-/-/tPA-/- mice do not develop the pervasive, multi-organ fibrin deposits, severe tissue damage, reduced fertility, and high morbidity and mortality observed in mice with a combined deficiency in tPA and the uPAR ligand, uPA. Furthermore, uPAR-/-/tPA-/- mice do not exhibit the profound impairment in wound repair seen in uPA-/-/tPA-/- mice when they are challenged with a full-thickness skin incision. These results indicate that plasminogen activation focused at the cell surface by uPAR is important in fibrin surveillance in the liver, but that uPA supplies sufficient fibrinolytic potential to clear fibrin deposits from most tissues and support wound healing without the benefit of either uPAR or tPA.