Though the electrostatic, ionic, van der Waals, Lennard-Jones, hydrogen bonding, and other forces play an important role in the energy function minimized at a protein's native state, it is widely believed that the hydrophobic force is the dominant term in protein folding. Here we attempt to quantify the extent to which the hydrophobic force determines the positions of the backbone alpha-carbon atoms in PDB data, by applying Monte-Carlo and genetic algorithms to determine the predicted conformation with minimum energy, where only the hydrophobic force is considered (i.e. Dill's HP-model, and refinements using Woese's polar requirement). This is done by computing the root mean square deviation between the normalized distance matrix D = (di,j) (di,j is normalized Euclidean distance between residues ri and rj) for PDB data with that obtained from the output of our algorithms. Our program was run on the database of ancient conserved regions drawn from GenBank 101 generously supplied by W. Gilbert's lab, as well as medium-sized proteins (E. Coli RecA, 2reb, Erythrocruorin, 1eca, and Actinidin 2act). The root mean square deviation (RMSD) between distance matrices derived from the PDB data and from our program output is quite small, and by comparison with RMSD between PDB data and random coils, allows a quantification of the hydrophobic force contribution. A preliminary version of this paper appeared at GCB'99 (http:¿bibiserv.techfak.uni-bielefeld.de/gcb9 9/).

We examine the role of residues of secondary structure in recognition of the native structure of proteins. The accuracy of recognition was estimated by computing the Z-score values for fragments of protein chains in threading tests. By testing different combinations of secondary structure fragments of 240 non-homologous proteins we show that the overwhelming majority of proteins can be successfully recognized by the energies of interaction between residues of secondary structure. We also found that beta-structures contribute more significantly to fold recognition than alpha-helices or loops. To validate the Z-score calculations in measuring the accuracy of recognition we evaluated the deviation of the energy distribution from the normal law. The normal law satisfactory approximates the shape of the energy distribution for the majority of proteins and chain fragments; however, deviations are often observed for short fragments and for fragments with relatively high Z-score values. The results of the study justify recognition of remote homologs by threading methods based on a backbone of secondary structure rather than of a whole chain because loops of homologs differ more significantly than strands and helices, and the contribution of loops in structure recognition is relatively small.

This paper describes a computational framework for cell biological modeling and simulation that is based on the mapping of experimental biochemical and electrophysiological data onto experimental images. The framework is designed to enable the construction of complex general models that encompass the general class of problems coupling reaction and diffusion.

A new method (particularly suited to the analysis of High Throughput Screening data) is presented for the determination of quantitative structure activity relationships. The method, termed "Binary QSAR," accepts binary activity measurements (e.g., pass/fail or active/inactive) and molecular descriptor vectors as input. A Bayesian inference technique is used to predict whether or not a new compound will be active or inactive. Experiments were conducted on a data set of 1947 molecules. The results show that the method exhibits high accuracy and is robust to measurement errors.

This paper describes a new approach to problem solving by splitting up problem component parts between software and hardware. Our main idea arises from the combination of two previously published works. The first one proposed a conceptual environment of concept modelling in which the machine and the human expert interact. The second one reported an algorithm based on reconfigurable hardware system which outperforms any kind of previously published genetic data base scanning hardware or algorithms. Here we show how efficient the interaction between the machine and the expert is when the concept modelling is based on reconfigurable hardware system. Their cooperation is thus achieved with an real time interaction speed. The designed system has been partially applied to the recognition of primate splice junctions sites in genetic sequences.

Finite H-systems with n test tubes are splicing systems of n test tubes over a common molecular alphabet, sigma, with a filter Fi [symbol: see text] sigma for each test tube. Initially, arbitrary many copies of molecules and enzymes (splicing rules) from a finite set of molecules and enzymes are given to the test tubes that produce new molecules by splicing and filtering. It is known that any formal language can be generated by a finite H-system with 9 test tubes and that the results of finite H-systems with 6 test tubes are unpredictable. Here we present a rather simple proof that the results of finite H-systems with only 3 test tubes are unpredictable and that 4 test tubes suffices to generate any formal language.

A strategy for rational design of targeted combinatorial libraries is described. The aim of this approach is to select a subset of available building blocks for the library synthesis that are most likely to be present in the active compounds. Building blocks that are used in the underlying combinatorial chemical reaction are randomly assembled to produce virtual combinatorial library compounds, which are represented by various chemical descriptors. Stochastic algorithms (simulated annealing, genetic algorithms, neural net methods) are used to search the potentially large structural space of virtual chemical libraries in order to identify compounds similar to lead compound(-s). The selection of a virtual molecule as a candidate for the targeted library is based either on its chemical similarity to a biologically active probe or on its biological activity predicted from a pre-constructed QSAR equation. Frequency analysis of building block composition of the selected virtual compounds identifies building blocks that can be used in combinatorial synthesis of chemical libraries with high similarity to the lead compound(-s). This method is applied to rational design of the library with bradykinin potentiating activity. Twenty eight bradykinin potentiating pentapeptides were used as a training set for the development of a QSAR equation, and, alternatively, two active pentapeptides, VEWAK and VKWAP, were used as probe molecules. In each case, the frequency distribution of amino acids in the top 100 peptides suggested by the method resembles the frequency distribution of amino acids found in the active peptides. The results obtained after GA optimization also compared favorably with those obtained by the exhaustive analysis of all possible 3.2 millions pentapeptides.

We describe the Object Technology Framework (OTF) software system developed at the University of California, San Francisco Computer Graphics Laboratory for creating C+2 classes that facilitate rapid biomolecular application development and the application of the OTF to collagen modeling. C+2 class libraries for accessing and manipulating data from standard scientific data sources can be generated using the program genlib and its class library toolkit Molecule, thereby facilitating development of new applications. Use of the OTF for generating ideal collagen structural models (gencollagen) is described. The source code for the OTF is freely available at http:/(/)www.cgl.ucsf.edu/off/ to interested application developers.

We present a global search technique for finding the global minimal conformation of a sequence in Dill's HP-lattice model. The HP-lattice model is a simplified model of proteins, that has become a major tool for investigating general properties of protein folding. The search technique uses constraint programming for efficiently pruning the search tree. We state the problem of structure prediction in the HP-lattice model and describe our implementation using the Oz-system.

We introduce an algorithm, LLLAMA, which combines simple pattern recognizers into a general method for estimating the entropy of a sequence. Each pattern recognizer exploits a partial match between subsequences to build a model of the sequence. Since the primary features of interest in biological sequence domains are subsequences with small variations in exact composition, LLLAMA is particularly suited to such domains. We describe two methods, LLLAMA-length and LLLAMA-alone, which use this entropy estimate to perform maximum a posteriori classification. We apply these methods to several problems in three-dimensional structure classification of short DNA sequences. The results include a surprisingly low 3.6% error rate in predicting helical conformation of oligonucleotides. We compare our results to those obtained using more traditional methods for automated generation of classifiers.