Bio-support vector machines for computational proteomics

MOTIVATION: One of the most important issues in computational proteomics is to produce a prediction model for the classification or annotation of biological function of novel protein sequences. In order to improve the prediction accuracy, much attention has been paid to the improvement of the performance of the algorithms used, few is for solving the fundamental issue, namely, amino acid encoding as most existing pattern recognition algorithms are unable to recognize amino acids in protein sequences. Importantly, the most commonly used amino acid encoding method has the flaw that leads to large computational cost and recognition bias. RESULTS: By replacing kernel functions of support vector machines (SVMs) with amino acid similarity measurement matrices, we have modified SVMs, a new type of pattern recognition algorithm for analysing protein sequences, particularly for proteolytic cleavage site prediction. We refer to the modified SVMs as bio-support vector machine. When applied to the prediction of HIV protease cleavage sites, the new method has shown a remarkable advantage in reducing the model complexity and enhancing the model robustness.     

Insights from modeling the tertiary structure of human BACE2

BACE1, or beta-secretase, is a putative prime therapeutic target for the treatment of Alzheimer's disease. Mapping to the Down syndrome critical region (chromosome 21) and identified as a homologue of BACE1, BACE2 also cleaves amyloid precursor protein at the beta-site. Thus, BACE2, named also as Asp1 or Memapsin1, represents a second beta-secretase candidate. In this paper, the tertiary structure of the protease domain of BACE2 was developed. Although the overall structural topology between BACE1 and BACE2 protease domains is quite similar, the former contains 3 disulfide bonds but the latter only two. Particularly, a subtle structural difference around the DTG/DSG active site between the two structures has been observed that is useful for the in-depth selectivity study of BACE1 and BACE2 inhibitors, stimulating new therapeutic strategies for the treatment of Alzheimer's disease and Down syndrome as well.     

'Signature sets', minimal fragment sets for identifying protein disulfide structures with cyanylation-based mass mapping methodology

Our cyanylation (CN)-based methodology for determining disulfide structure of cystinyl proteins overcomes the limitations of conventional proteolytic methods. However, the CN-based method has the potential drawback that occasionally some CN-induced cleavage fragments may not be detected. We show that CN-based methods can overcome the failure to detect fragments by demonstrating the existence of small 'signature sets' of fragments. The link between signature sets and the robustness of CN-based methodology is validated by two case studies.     

Gas chromatography-isotope dilution mass spectrometry preceded by liquid-liquid extraction and chemical derivatization for the determination of ketamine and norketamine in urine

An analytical scheme using gas chromatography (GC)-isotope dilution mass spectrometry (MS) assisted by precedent liquid-liquid extraction (LLE) and chemical derivatization (ChD) is described for the simultaneous determination of ketamine (KT) and its major metabolite, norketamine (NK), in urine. The simultaneous ChD of the two analytes, one primary amine and one secondary amine, with pentafluorobenzoyl chloride (PFBC) has not only enhanced their instrumental responses and mass-spectrum uniqueness but also afforded more proper yet easier selection of qualifier and quantifier ions and hence achieved more evidential identification and quantitation. Thus, the regression calibration curves for KT and NK in urine are linear within 100-5000 ng/ml, with correlation coefficients typically exceeding 0.99 and NK curves exclusively showing larger slopes than KT curves. The method limits of detection (LODs) determined by two definitions for KT and NK range from 3 to 75 ng/ml, and limits of quantitation (LOQs) from 9 to 100 ng/ml. The mean recoveries (N = 3) calculated for the LLE/ChD of KT and NK from 50 and 100 microl, respectively, of a 100 microg/ml urinary spike vary from 71.0 to 97.8%, with NK consistently giving better recoveries than KT. The precisions (N = 3) calculated for the total analyses of four real-case samples are typically below 12.3%. GC-MS operated in the positive ion chemical ionization (PCI) mode can offer both qualitative and quantitative information complementary to those given by the EI mode. The proposed scheme is simple, effective, reliable, and robust. It may serve as a confirmatory protocol for forensic urine drug testing.     

Adaptive combinatorial design to explore large experimental spaces: approach and validation

Systems biology requires mathematical tools not only to analyse large genomic datasets, but also to explore large experimental spaces in a systematic yet economical way. We demonstrate that two-factor combinatorial design (CD), shown to be useful in software testing, can be used to design a small set of experiments that would allow biologists to explore larger experimental spaces. Further, the results of an initial set of experiments can be used to seed further 'Adaptive' CD experimental designs. As a proof of principle, we demonstrate the usefulness of this Adaptive CD approach by analysing data from the effects of six binary inputs on the regulation of genes in the N-assimilation pathway of Arabidopsis. This CD approach identified the more important regulatory signals previously discovered by traditional experiments using far fewer experiments, and also identified examples of input interactions previously unknown. Tests using simulated data show that Adaptive CD suffers from fewer false positives than traditional experimental designs in determining decisive inputs, and succeeds far more often than traditional or random experimental designs in determining when genes are regulated by input interactions. We conclude that Adaptive CD offers an economical framework for discovering dominant inputs and interactions that affect different aspects of genomic outputs and organismal responses.     

LUCY2: an interactive DNA sequence quality trimming and vector removal tool

Lucy2 is a raw DNA sequence trimming and visualization tool based on the popular command-line Lucy1. Users can change parameters, trim multiple sequences and visualize the results within an integrated, easy-to-use graphical user interface. Lucy2 is designed specifically for non-programmers to use, and is currently available on Windows, Linux and MacOS X. Source code is also available for porting to the other platforms. AVAILABILITY: Lucy2 is distributed under the GNU General Public License and can be downloaded from www.complex.iastate.edu     

Picky: oligo microarray design for large genomes

MOTIVATION: Many large genomes are getting sequenced nowadays. Biologists are eager to start microarray analysis taking advantage of all known genes of a species, but existing microarray design tools were very inefficient for large genomes. Also, many existing tools operate in a batch mode that does not assure best designs. RESULTS: Picky is an efficient oligo microarray design tool for large genomes. Picky integrates novel computer science techniques and the best known nearest-neighbor parameters to quickly identify sequence similarities and estimate their hybridization properties. Oligos designed by Picky are computationally optimized to guarantee the best specificity, sensitivity and uniformity under the given design constrains. Picky can be used to design arrays for whole genomes, or for only a subset of genes. The latter can still be screened against a whole genome to attain the same quality as a whole genome array, thereby permitting low budget, pathway-specific experiments to be conducted with large genomes. Picky is the fastest oligo array design tool currently available to the public, requiring only a few hours to process large gene sets from rice, maize or human.     

Neuregulin 1 and susceptibility to schizophrenia

The cause of schizophrenia is unknown, but it has a significant genetic component. Pharmacologic studies, studies of gene expression in man, and studies of mouse mutants suggest involvement of glutamate and dopamine neurotransmitter systems. However, so far, strong association has not been found between schizophrenia and variants of the genes encoding components of these systems. Here, we report the results of a genomewide scan of schizophrenia families in Iceland; these results support previous work, done in five populations, showing that schizophrenia maps to chromosome 8p. Extensive fine-mapping of the 8p locus and haplotype-association analysis, supplemented by a transmission/disequilibrium test, identifies neuregulin 1 (NRG1) as a candidate gene for schizophrenia. NRG1 is expressed at central nervous system synapses and has a clear role in the expression and activation of neurotransmitter receptors, including glutamate receptors. Mutant mice heterozygous for either NRG1 or its receptor, ErbB4, show a behavioral phenotype that overlaps with mouse models for schizophrenia. Furthermore, NRG1 hypomorphs have fewer functional NMDA receptors than wild-type mice. We also demonstrate that the behavioral phenotypes of the NRG1 hypomorphs are partially reversible with clozapine, an atypical antipsychotic drug used to treat schizophrenia.     

Prediction of the tertiary structure of the beta-secretase zymogen

beta-Secretase, also known as BACE, is a transmembrane aspartyl protease, which generates the N terminus of Alzheimer's disease amyloid beta-peptide. The activity of beta-secretase is the rate-limiting step of brain plaques production in vivo, and hence is a potential target for disease modifying drugs for Alzheimer's disease. To better understand the mechanism of action of beta-secretase and help explore novel strategies for drug discovery for Alzheimer's disease, it is important to elucidate the three-dimensional structure of its zymogen. Based on the X-ray structure of the enzyme's protease domain and the X-ray structure of pepsinogen, a model of the three-dimensional structure of the beta-secretase zymogen has been constructed. Comparison of the computed structure of pro-BACE with X-ray structures of pepsinogen and progastricsin (two other pro-aspartyl proteases) reveals a significant difference in the relationship of the pro-segment to the catalytic aspartates. In both pepsinogen and progastricsin a lysine side-chain in the pro-segment forms a salt bridge to the two catalytic aspartates, occupying the position normally occupied by a catalytic water. In the pro-BACE model there is no salt bridge, and the corresponding residue-a proline-does not interact at all with the catalytic residues. These findings can be used to elucidate the recent observations that the pro-domain of beta-secretase does not suppress activity as in a strict zymogen but does appear to facilitate proper folding of an active protease domain. The predicted three-dimensional structure of beta-secretase zymogen and the relevant findings might also provide useful insights for rational design of effective drugs against Alzheimer's disease.     

Using subsite coupling to predict signal peptides

Given a nascent protein sequence, how can one predict its signal peptide or "Zipcode" sequence? This is a first important problem for scientists to use signal peptides as a vehicle to find new drugs or to reprogram cells for gene therapy. Based on a model that takes into account the coupling effect among some key subsites, the so-called [-3, -1, +1] coupling model, a new prediction algorithm is developed. The overall rate of correct prediction for 1939 secretory proteins and 1440 non-secretary proteins was over 92%. It has not escaped our attention that the new method may also serve as a useful tool for helping investigate further many unclear details regarding the molecular mechanism of the ZIP code protein-sorting system in cells.