ACD/Labs unveils a new version of its NMR prediction software building on large databases of experimental data, introducing new structural diversity, and continually developing proprietary algorithms
ACD/Labs says that, as a result of its continuing development of its software, the accuracy of NMR prediction for all nuclei treated by ACD/Labs has reached an all-time best and is approaching statistical benchmarks that have taken over ten years to achieve.
The algorithmic prediction of accurate NMR chemical shifts is a challenging yet appropriate venture due to the benefits that could be delivered to the chemistry community.
Advanced Chemistry Development (ACD/Labs) says it has long pursued the challenge of 'optimal performance' in its NMR predictors, and during this journey has continued to expand the content databases of reviewed data and extended its algorithms to improve the performance of its products.
"NMR prediction is only a means to an end - either using predicted spectra to verify a structural hypothesis, or to assist in the process of structure elucidation," says Antony Williams, vice president and chief science officer at ACD/Labs.
"ACD/Labs has been working on the four pillars of NMR software for over a decade - processing, prediction, structure verification, and computer-assisted structure elucidation (Case).
"Since structure verification and Case are highly dependent on the quality of NMR prediction, our efforts balance a focus on prediction with the target objectives of verification and elucidation.
"We remain dedicated to delivering on these objectives as demonstrated by our annual improvements in prediction accuracy".
Until now, ACD/Labs has persisted on the delivery of a modified Hose (hierarchial organisation of spherical environments) code approach for NMR prediction.
The reason for this persistence has been the ongoing demonstration of better predictions of Hose code over neural networks (NN) from numerous internal validation studies carried out over the years.
Nevertheless, the massive speed advantages of NN calculations (100s to 1000s of times faster) prompted the development of neural network-based predictions to support specific needs in ACD/Labs's Structure Elucidator application.
During the research and development stages of this process it became surprisingly clear that the NN algorithm was outperforming the Hose code algorithm previously available in version 9.
Furthermore, the information gathered from this development work helped to improve and optimise the version 10 Hose code algorithms.
The final results from these studies resulted in the observation that version 10 Hose code predictions still outperform NN predictions in all criteria except speed.
"We are very pleased with the latest improvements in our prediction accuracy," comments Brent Lefebvre, NMR product manager, ACD/Labs.
"We are always comparing our new version results with previous versions, and with other competitive products in the marketplace.
"We are confident that our prediction accuracy remains unsurpassed in the marketplace and we are excited about the implementation of a new neural network algorithm.
"Considering this was our first attempt at the development and implementation of a neural network algorithm, we believe that this opens up even more prosperous opportunities to potentially combine the strength of both algorithms in an intelligent way as we continually strive to produce the most accurate NMR predictions possible.
"The validation of algorithms is made annually by testing the prediction performance against a large test dataset of 1H and 13C NMR chemical shifts and their associated published structures.
"The study showed that the calculated standard errors for 1H and 13C chemical shift prediction accuracy in version 10 are 0.22 and 2.11ppm, respectively", adds Williams.
"Ten years ago, standard errors of 0.3 and 3.0ppm for 1H and 13C NMR chemical shift prediction were appropriate and sufficient in the majority of cases.
"However, as we passed that benchmark and focused on the challenges of assisted verification and elucidation, we set our targets on values of 0.2 and 2ppm.
"With both innovative and incremental improvements, we expect to deliver on these goals in the very near future."
