Mapping the human proteome for non-redundant peptide islands.

We describe immune-proteome structures using libraries of protein fragments that define a structural immunological alphabet. We propose and validate such an alphabet as i) composed of letters of five consecutive amino acids, pentapeptide units being sufficient minimal antigenic determinants in a protein, and ii) characterized by low-similarity to human proteins, so representing structures unknown to the host and potentially able to evoke an immune response. In this context, we have thoroughly sifted through the entire human proteome searching for non-redundant protein motifs. Here, for the first time, a complete sequence redundancy dissection of the human proteome has been conducted. The non-redundant peptide islands in the human proteome have been quantified and catalogued according to the amino acid length. The library of uniquely occurring n-peptide sequences that was obtained is characterized by a logarithmic decrease of the number of non-redundant peptides as a function of the peptide length. This library represents a highly specific catalogue of molecular protein signatures, the possible use of which in cancer/autoimmunity research is discussed, with a major focus on non-redundant dodecamer sequences.

Sub-epitopic dissection of HCV E1315-328HRMAWDMMMNWSPT sequence by similarity analysis.

Our labs are focused on identifying amino acid sequences having the ability to react specifically with the functional binding site of a complementary antibody. Our epitopic definition is based on the analysis of the similarity level of antigenic amino acid sequences to the host proteome. Here, the similarity profile to the human proteome of an HCV E1 immunodominant epitope, i.e. the HCV E1(315-328)HRMAWDMMMNWSPT sequence, led to i) characterizing the immunoreactive HCV E1 315-328 region as a sequence endowed with a low level of similarity to human proteins; ii) defining 2 contiguous immunodominant linear determinants respectively located at the NH(2) and COOH terminus of the conserved viral antigenic sequence. This study supports the hypothesis that low sequence similarity to the host's proteome modulates the pool of epitopic amino acid sequences in a viral antigen, and appears of potential value in defining immunogenic viral peptide sequences to be used in immunotherapeutic approaches for HCV treatment.

Peptidology: short amino acid modules in cell biology and immunology.

Short amino acid motifs, either linear sequences or discontinuous amino acid groupings, can interact with specific protein domains, so exerting a central role in cell adhesion, signal transduction, hormone activity, regulation of transcript expression, enzyme activity, and antigen-antibody interaction. Here, we analyze the literature for such critical short amino acid motifs to determine the minimal peptide length involved in biologically important interactions. We report the pentapeptide unit as a common minimal amino acid sequence critically involved in peptide-protein interaction and immune recognition. The present survey may have implications in defining the dimensional module for peptide-based therapeutical approaches such as the development of novel antibiotics, enzyme inhibitors/activators, mimetic agonists/antagonists of neuropeptides, thrombolitic agents, specific anti-viral agents, etc. In such a therapeutical context, it is of considerable interest that low molecular weight peptides can easily cross biological barriers, are less susceptible to protease attacks, and can be administered at high concentrations. In addition, small peptides are a rational target for strategies aimed at antigen-specific immunotherapeutical intervention. As an example, specific short peptide fragments might be used to elicit antibodies capable of reacting with the full-length proteins containing the peptide fragment's amino acid sequence, so abolishing the risk of cross-reactivity.

Modeling gene expression from microarray expression data with state-space equations.

We describe a new method to model gene expression from time-course gene expression data. The modelling is in terms of state-space descriptions of linear systems. A cell can be considered to be a system where the behaviours (responses) of the cell depend completely on the current internal state plus any external inputs. The gene expression levels in the cell provide information about the behaviours of the cell. In previously proposed methods, genes were viewed as internal state variables of a cellular system and their expression levels were the values of the intemal state variables. This viewpoint has suffered from the underestimation of the model parameters. Instead, we view genes as the observation variables, whose expression values depend on the current intemal state variables and any external input. Factor analysis is used to identify the internal state variables, and Bayesian Information Criterion (BIC) is used to determine the number of the internal state variables. By building dynamic equations of the internal state variables and the relationships between the internal state variables and the observation variables (gene expression profiles), we get state-space descriptions of gene expression model. In the present method, model parameters may be unambiguously identified from time-course gene expression data. We apply the method to two time-course gene expression datasets to illustrate it.

MolCom: a method to compare protein molecules based on 3-D structural and chemical similarity.

This paper describes an improved method for conducting global feature comparisons of protein molecules in three dimensions and for producing a new form of multiple structure alignment. Our automated MolCom method incorporates an octtree strategy to partition and examine molecular properties in three-dimensional space at multiple levels of analysis. The MolCom method's multiple alignment is in the form of an octtree which locates regions in three-dimensional space where correspondence between molecules is identified based on a dynamic set of molecular features. MolCom offers a practical solution to the inherent compromise between computational complexity and analytical detail. MolCom is currently the only method that can analyze and compare a series of defined physicochemical properties using multiple, simultaneous levels of resolution. It is also the only method that provides a consensus structure outlining precisely where the similarity exists in three-dimensional space. Using a modest-sized collection of structural properties, separate experiments were conducted to calibrate MolCom and to verify that the spatial analyses and resulting structure alignments accurately identified both similar and dissimilar structures. The accuracy of MolCom was found to be over 99% and the similarity scores correlated strongly with the z-scores of the Alignment by Incremental Combinatorial Extension of the Optimal Path method.

Application of a multi-structure neural network (MSNN) to sorting pistachio nuts.

A multi-structure neural network (MSNN) classifier consisting of four discriminators followed by a maximum selector was designed and applied to classification of four grades of pistachio nuts. Each discriminator was a multi-layer feed-forward neural network with two hidden layers and a single-neuron output layer. Fourier descriptors of the nuts' boundaries and their area were used as the recognition features. The individual discriminators were trained using a biased technique and a back-propagation algorithm. The MSNN classifier gave an average classification performance of 95.0%. This was an increase of 14.8% over the performance of a multi-layer neural network (MLNN) with similar complexity for classifying the same set of patterns.