A NEURAL-NETWORK APPROACH TO THE RAPID COMPUTATION OF ROTATIONAL CORRELATION TIMES FROM SLOW MOTIONAL ESR-SPECTRA

We explore the use of feed forward artificial neural networks for dete rmining rotational correlation times from slow motional nitroxide elec tron spin resonance spectra. This approach is rapid and potentially el iminates the need for traditional iterative fitting procedures. Two ne tworks are examined: the radial basis network and the multilayer perce ptron. Although the radial basis network trains rapidly and performs w ell on simulated spectra, it is less satisfactory when applied to expe rimental spectra. In contrast, the multilayer perceptron trains slowly but is excellent at extracting correlation times from experimental sp ectra. In addition, the multilayer perceptron operates well in the pre sence of noise as long as the signal-to-noise ratio is greater than ap proximately 200/1. These findings suggest neural networks offer a prom ising approach for rapidly extracting correlation times without the ne ed for iterative simulations. (C) 1998 Academic Press.