The Automatic Structure Verification (ASV) project of Mestrelab is progressing quite nicely, even though some Users are impatient and, understandably, pressing for more and sooner). The development process is a complex uphill fight since we want a fully automatic artificial intelligence (AI) working with real spectra, not just some partially computer-assisted approach requiring extremely high quality, manually pre-edited spectra. Nothing so sweeping has ever been attempted before (and not even contemplated) and yet our present system has already reached the state when it is really useful in actual everyday practice
The ASV feature is configured into ASV tasks which, in turn, are composed of multiple ASV tests. Its development consists in devising novel ASV tests embodying expert NMR knowledge, and combining them together within a modern scoring system. This implies the presence of a number of numeric parameters which are then tuned to an extensive set of well characterized experimental data sets (benchmarks).
The overall ASV structure is universal enough to accommodate both 1D and 2D NMR spectra and even merge them with results of other techniques such as MS.
We have developed a number of 1D NMR tests and an HSQC test. These are all based on the experimental spectra, the assumed molecule, and the predicted NMR parameters, but focus on different aspects of the predicted-versus-experimental comparison. They work very nicely; in fact, the major obstacles we have encountered so far did not regard the tests themselves but almost exclusively apparently unrelated aspects such as the quality of the experimental spectra, solvent recognition, and labile peaks recognition.
The ASV wizard has now got as far as one can go (mathematically speaking) without attempting actual assignment of individual nuclei to spectral peaks or multiplets. This will be the next step of the whole long-term development.