Andrew T. Brint

Algorithms for the identifications of three-dimensional maximal common substructures

Two algorithms are described for the identification of the maximal substructures common to two (or more) three-dimensional chemical structures, where a substructure consists of a set of atoms and the associated interatomic distances. The algorithm of Crandell and Smith involves a breadth-first tree search procedure in which substructures are expanded as they are shown to be common to all of the molecules under consideration. The clique-detection algorithm involves the identification of cliques in the correspondence graph linking matching atoms and interatomic distances in pairs of structures that are being compared. This second algorithm is shown to be substantially faster in operation than the Crandell and Smith algorithm when applied to structures taken from the Cambridge Crystallographic Data Bank, and an extension to the algorithm is described that allows it to be used for the identification of the maximal substructure common to arbitrary numbers of molecules.

Pharmacophoric pattern matching in files of 3d chemical structures: comparison of geometric searching algorithms

Abstract This paper reports a comparative evaluation of four algorithms that can be used to determine the presence or absence of a set of atoms and associated interatomic distances in 3D chemical structures. The geometric searching algorithms tested are that described by Lesk for the identification of patterns in proteins, one derived from the set reduction techniques used for substructure searching in files of 2D chemical structures, a procedure based on clique detection, and Ullmans subgraph isomorphism algorithm. Tests with structures from the Cambridge Crystallographic Data Bank demonstrate the general superiority of Ullmans algorithm for data of this sort.

Upperbound procedures for the identification of similar three-dimensional chemical structures.

SummaryThis paper describes techniques for calculating the degree of similarity between an input query molecule and each of the molecules in a database of 3-D chemical structures. The inter-molecular similarity measure used is the number of atoms in the 3-D common substructure (CS) between the two molecules which are being compared. The identification of 3-D CSs is very demanding of computational resources, even when an efficient clique detection algorithm is used for this purpose. Two types of upperbound calculation are described which allow reductions in the number of exact CS searches which need to be carried out to identify those molecules from a database which are similar to a 3-D target molecule.

Rapid geometric searching in protein structures

Abstract This paper discusses the implementation of geometric searching in protein structures. A comparison of geometric searching algorithms that have been used for substructure searching in small 3D molecules suggests that the algorithm due to Lesk is most appropriate for searching for patterns of atoms in proteins; however, the computational requirements of this algorithm are considerable. An improved, two-stage procedure is described in which the Lesk algorithm is used as a precursor to the subgraph isomorphism algorithm of Ullman, and it is demonstrated that geometric searching can be implemented with a reasonable degree of efficiency.

Identifying 3D maximal common substructures using Transputer networks

Abstract This paper describes the use of a multiprocessor system for identifying the maximal common substructure (MCS) between pairs of three-dimensional (3D) chemical structures. The system is constructed from Transputers, 32-bit RISC microprocessors produced by Inmos Ltd., linked together in a tree network. The MCS algorithm used, developed by Crandell and Smith, identifies the MCS by a breadth-first search in which individual atoms common to the structures are extended one atom at a time until no further extension of the common substructure can be obtained. Experiments using a Pascal-based simulation package demonstrate the feasibility of using a multiprocessor system to increase the speed of MCS identification. Experiments with networks of Transputers demonstrate that substantial increases in speed can be achieved in practice if, and only if, the MCS is large.

Predicting a house's selling price through inflating its previous selling price

This paper considers how accurately inflating the previous selling price of a modern property predicts its selling price. Predicting a houses value is an important capability as it allows how the asking price affects the time to sale and the price achieved, to be modelled. The analysis is carried out on a data set of 105 pairs of earlier and later selling prices for UK properties constructed since January 1999. As an alternative to using published house price indices for inflating the prices, a novel approach for modifying the published house price indices through the use of observed repeat sales of properties is put forward and analysed. Using the best published index gives an average predictive error of 10.9% while using the published index modified by repeat-sales information, gives an average predictive error of 8.4%.

Chemical graph matching using transputer networks

Abstract This paper discusses the use of networks of transputers for the matching of the labelled graphs which are used to represent chemical structures in computer-based chemical information systems; in particular, the implementation of a relaxation algorithm for chemical substructure searching is described. Tests with a doubly-linked chain of transputers suggest that near-linear speedups can be obtained by a partitioning of the database which is to be searched; lesser speedups are obtained with other network configurations. Current work is described involving the exploitation of an alternative level of parallelism in the relaxation algorithm and the parallel implementation of an algorithm for the identification of the maximal substructures common to a pair of chemical compounds.

Substructure Searching in Files of Three-Dimensional Chemical Structures

This paper discusses techniques for chemical substructure searching in files of three-dimensional (3-D) chemical structures. A methodology is presented for the selection of search screens which are based upon interatomic distance information, the use of these screens on a file of 3-D chemical structures taken from the Cambridge Crystallographic Data Bank (CCDB) shows that they allow searches for typical pharmacophoric patterns to be carried out with high efficiency. A range of subgraph isomorphism algorithms is described for geometric searching, the 3-D equivalent of atom-by-atom searching in conventional substructure search systems. Current research involves the development of hardware and software techniques for the identification of the maximal substructures common to sets of 3-D structures, and the extension of the work on substructure searching in the CCDB to the macromolecular structures in the Protein Data Bank.

Improving estimates of asset condition using historical data

Regularly updating the estimated conditions of the asset base is important when managing infrastructure networks. This is usually done using a sampling programme in which some or all of the assets are inspected over a specified time horizon. Commonly, only a proportion of assets are inspected in each year. Therefore, the asset managers need to be able to use this updated but partial knowledge to get the best possible view of the condition of the whole asset base. This paper considers approaches to doing this in the situation where the asset conditions are ordinal, meaning that the condition measurements fall into discrete ordered categories. The performances of several straightforward algorithms are analysed in terms of how good the predictions are. It is concluded that an ordinal logistic approach gives the best results, but a linear regression model gives acceptable results and has the advantage of being easier to implement.

Using grouped smart meter data in phase identification

Abstract Access to smart meter data will enable electricity distribution companies to have a far clearer picture of the operation of their low voltage networks. This in turn will assist in the more active management of these networks. An important current knowledge gap is knowing for certain which phase each customer is connected to. Matching the loads from the smart meter with the loads measured on different phases at the substation has the capability to fill this gap. However, in the United Kingdom at the half hourly level only the loads from groups of meters will be available to the network operators. Therefore, a method is described for using this grouped data to assist with determining each customers phase when the phase of most meters is correctly known. The method is analysed using the load readings from a data set of 96 smart meters. It successfully ranks the mixed phase groups very highly compared with the single phase groups.