Sándor Szalma

A hidden Markov model with molecular mechanics energy-scoring function for transmembrane helix prediction

A range of methods has been developed to predict transmembrane helices and their topologies. Although most of these algorithms give good predictions, no single method consistently outperforms the others. However, combining different algorithms is one approach that can potentially improve the accuracy of the prediction. We developed a new method that initially uses a hidden Markov model to predict alternative models for membrane spanning helices in proteins. The algorithm subsequently identifies the best among models by ranking them using a novel scoring function based on the folding energy of transmembrane helical fragments. This folding of helical fragments and the incorporation into membrane is modeled using CHARMm, extended with the Generalized Born surface area solvent model (GBSA/IM) with implicit membrane. The combined method reported here, TMHGB significantly increases the accuracy of the original hidden Markov model-based algorithm.

Assessment of putative protein targets derived from the SARS genome.

The ability to rapidly and reliably develop hypotheses on the function of newly discovered protein sequences requires systematic and comprehensive analysis. Such an analysis, embodied within the DS GeneAtlas™ pipeline, has been used to critically evaluate the severe acute respiratory syndrome (SARS) genome with the goal of identifying new potential targets for viral therapeutic intervention. This paper discusses several new functional hypotheses on the roles played by the constituent gene products of SARS, and will serve as an example of how such assignments can be developed or extended on other systems of interest.

Target validation through high throughput proteomics analysis

Abstract High throughput functional annotation of the proteome has emerged as a standard tool for target identification. In contrast, target validation, which requires detailed analysis of biological function, has until recently remained an essentially experimental low throughput activity. Currently, there is considerable interest in accelerating and improving the validation process to counter the declining number of small-molecule-based therapeutics being released onto the market. Progress in high throughput proteomics is a key technology in this respect. Uniquely, it offers the ability to rapidly identify and characterize networks of interacting proteins, which in turn presents new opportunities to develop alternative lead development strategies.

FAUST: An Algorithm for Extracting Functionally Relevant Templates from Protein Structures

FAUST (Functional Annotations Using Structural Templates) is an algorithm for: extraction of functionally relevant templates from protein structures and using such templates to annotate novel structures. Proteins and structural templates are represented as colored, undirected graphs with atoms as nodes and interatomic distances as edge weights. Node colors are based on chemical identities of atoms. Edge labels are equivalent if interatomic distances for corresponding nodes (atoms) differ less than a threshold value. We define FAUST structural template as a common subgraph of a set of graphs corresponding to two or more functionally related proteins. Pairs of functionally related protein structures are searched for sets of chemically equivalent atoms whose interatomic distances are conserved in both structures. Structural templates resulting from such pair wise searches are then combined to maximize classification performance on a training set of irredundant protein structures. The resulting structural template provides new language for description of structure--function relationship in proteins. These templates are used for active and binding site identification in protein structures. We are demonstrating here structural template extraction results for the highly divergent family of serine proteases. We compare FAUST templates to the standard description of the serine proteases active site pattern conservation and demonstrate depth of information captured in such description. Also, we present preliminary results of the high-throughput protein structure database annotations with a comprehensive library of FAUST templates.