Premium
Performance analysis of the double‐iterated Kalman filter for molecular structure estimation
Author(s) -
Delfini D.,
Nicolini C.,
Carrara E. A.
Publication year - 1996
Publication title -
journal of computational chemistry
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.907
H-Index - 188
eISSN - 1096-987X
pISSN - 0192-8651
DOI - 10.1002/(sici)1096-987x(19960115)17:1<74::aid-jcc7>3.0.co;2-y
Subject(s) - iterated function , covariance , algorithm , extended kalman filter , computer science , estimator , kalman filter , covariance matrix , filter (signal processing) , mathematical optimization , mathematics , artificial intelligence , mathematical analysis , statistics , computer vision
A possible application of a novel double‐iterated Kalman filter (DIKF) as an algorithm for molecular structure determination is investigated in this work. Unlike traditional optimization algorithms, the DIKF does not exploit experimental nuclear magnetic resonance (NMR) constraints in a penalty function to be minimized but used them to filter the atomic coordinates. Furthermore, it is a nonlinear Bayesian estimator able to handle the uncertainty in the experimental data and in the computed structures, represented as covariance matrices. The algorithm presented applies all constraints simultaneously, in contrast with DIKF algorithms for structure determination found in literature, which apply the constraints one at a time. The performances of both paradigms are tested and compared with those obtained by a commonly used optimization algorithm (based on the conjugate gradient method). Besides providing estimates of the conformational uncertainty directly in the final covariance matrix, DIKF algorithms appear to generate structures with a better stereochemistry and be able to work with realistically imprecise constraints, while time performances are strongly affected by the heavy matricial calculations they require. © 1996 by John Wiley & Sons, Inc.