An approximate orthogonalization method called "pseudo-orthogonalization" was derived here from the Schmidt orthogonalization in a partial space. Precision of the orthogonalization is decimal 2 digits when the partial space of 1/3 of the Schmidt orthogonalization is used. The speed-up ratio is 3.3 with 1000 vectors of 1500th dimension. Although the pseudo-orthogonalization is high-speed, the precision is inadequate for general numerical calculations. Two possible approaches to improve this precision were considered: (a) introducing a band structure in the vectors, and (b) restricting the sign of the vector elements. These approaches improved the precision by a factor of 2.1 and 1.5, respectively, without increasing the CPU time. To demonstrate the applicability of this high-speed pseudo-orthogonalization for large-scale calculations, we coded a complex-form of it into the SCF part of the Car-Parrinello method. In calculations of the total energy for bulk-silicon, the difference in total energy calculated using the pseudo-orthogonalization and that using the Schmidt orthogonalization was less than O(-5) [a.u.]. Based on this accuracy, this pseudo-orthogonalization can be used to speed-up the Car-Parrinello method when it is applied to large-scale calculations.
Many efforts have been devoted to the development of computer-aided prediction of drug toxicity over the decades, but at present, systems and programs available for predicting teratogenicity from chemical structures do not always give satisfactory answers yet, mainly because of the complex and unknown mechanism of reproductive and developmental toxicity. We developed a novel algorithm and implemented in the program "SimScore" to evaluate quantitatively the structural similarity score of a target compound with the teratogenic drugs which are defined as serious human teratogens by the United States Food and Drug Administration. In SimScore, a molecular structure is divided into its skeletal and substituent parts in order to perform similarity comparison for these parts independently. This idea is based on that compounds with the same or similar skeleton show a similar biological activity, but their activity strengths depend on the variation of substituents. We demonstrated the usefulness of SimScore by applying it to an example. SimScore will be used in our web-based information system about teratogenicity to predict the potential risk of query compounds.
MagSaki(A) software was developed for the magnetic analysis of dinuclear high-spin cobalt(II) complexes. This software performs magnetic analysis to determine magnetic parameters using five types of theoretical susceptibility equations. A characteristic feature of the software is that the exchange interaction can be treated anisotropically whereas the previous MagSaki software treats the exchange interaction isotropically.
A model for purification mechanisms in a river was proposed to express changes in the values of BOD (Biochemical Oxygen Demand), COD (Chemical Oxygen Demand), T-N (Total Nitrogen), and T-P (Total Phosphorous) as the combination of inflows, streams, and weirs. Because expressions for these functions are unknown, the model first constructs multi-layer neural-network functions based on observations, and then uses the derivatives to evaluate the cause and effect of pollution. For data of the Tamagawa through Tokyo, Japan in 2002, the model suggested that the cause of pollution is inflows from sewage, the stream of the Tamagawa has purification functions for COD and T-P, but has little ability for T-N, and the weirs have purification for COD, but have no purification for T-N and T-P. The results from the model were consistent with common sense for the water quality, and thus, there was no failure in the model.