by Steve Hayward, Technical Marketing Specialist, (ACD/Labs)
In our previous post, Andrew explored the challenges of empowering organizations to leverage their analytical science assets, which was one of the key reasons he re-joined the ACD/Labs family. While these challenges are at the core of our developmental targets, I’d like to take a few moments to explore one technology which is the underlying heart of our solutions for scientists involved in NMR.
From basic structure-to-spectrum consistency checks in ACD/Spectrus Processor, to complete elucidations of unknown structures, to fully-automated structure verification systems, the algorithms of ACD/NMR Predictors are perhaps now the unsung heroes of our software. Indeed, they were the first software we ever sold all the way back in 1994. But after becoming ubiquitous in our products (and that of other companies) is there any way to estimate how useful it is to have access to accurate NMR predictions? Besides helping university students with their assignments, that is…..
The benefits of structural assignment assistance, in one form or another, have been generally accepted for some time. At the same time, the evidence of incorrect structural assignments is abundant, and as detailed by Csaba Szantay in his excellent book, is often due to unwitting subjective analysis on the part of researchers. However, a user can flag structural assignment inconsistencies with appropriate guidance from a predicted spectrum, helping to avoid these initial mistakes. A powerful and useful NMR prediction algorithm must attain a high prediction accuracy within a short prediction time, something demonstrated by our software previously. However they are implemented, having access to NMR spectral predictions seems like common sense for any spectroscopist.
Last week also saw the release of a paper from Google scientists which postulated that their radical D-Wave 2X quantum computer had managed to achieve quantum annealing, a key stepping stone to the wider application of quantum computing. Essentially, this would allow the D-Wave 2X to perform up to ~108 times faster calculations compared to a classical computer, which may become useful for neural net predictive processing relying on exceptionally large databases.
However, of how much importance is this technology in the realm of NMR prediction and interpretation, especially given the advancements in neural net predictions already running on classical computers? After all, just this year neural net systems have been utilized everywhere from a predictive-text keyboard for mobile devices, to an early attempt at writing content for a tech blog. Neural net NMR prediction algorithms have been a part of our software since ~2008 (along with their inherent speed), with a weighted average between neural net and HOSE-code predictions now presented by default. So how large would a database of chemical shifts and coupling constants need to become before quantum computing would make an impact?
Perhaps at the moment the question is somewhat rhetorical; we already have the processing power to accomplish most of our desired processing and interpretative tasks. At the same time, a recent paper by Andres Castillo et al. demonstrates how NMR prediction algorithms can be evaluated, and potentially improved, without the need for manually-assigned experimental reference spectra, opening the door to automated analysis using very large datasets. It is conceivable that in the future, such analysis would require increased computing power—it seems we are always destined to produce ever-more powerful computers, and invent new technologies which require that power, at roughly the same pace.
As Hartmut Neven of Google points out (in this excellent article from Wired), we may look back at the D-Wave 2X’s accomplishments in the same way we view the Wright brothers’ first flight across a North Carolina beach. Perhaps, if quantum computing is the way towards artificial intelligence, with appropriate training and databases a computer system would be able to perform all the steps in an NMR experiment—capturing a spectrum, interpreting it, postulating and evaluating a structure, and preparing a report in prose describing the results for human evaluation. Only time will tell….