Predictive Methods for Organic Spectral Data Simulation

Abstract

New chemical entities (NCE) with potential bioactivity are synthesized, isolated, and thoroughly characterized for structure elucidation and purity before being subjected to further research. Spectroscopy is one of the most powerful means to deduce the correct structure and configuration of a compound or a fragment. In organic synthesis, the compounds are usually characterized by the spectral techniques such as ultraviolet–visible (UV–Vis), nuclear magnetic resonance (NMR), infrared (IR), mass spectrometry (MS), X-ray, etc. NMR and MS methods are employed in fragment-based drug discovery approaches to identify compounds from a high-throughput screen or a proteomics experiment. However, it is not possible to manually interpret the complex spectral data that require sophisticated computational tools for characterization. These tools aid in spectra analysis, peaks assignment, intensity, etc. and thereby annotate the compound with the appropriate functional group and fragments. The prediction algorithms are developed based on principles of quantum chemistry, machine learning, or simple database/pattern match-based methods. Some of the methods using quantum chemistry are accurate; however, they require more computational time; on the other hand, the machine learning methods such as neural network are faster but require more experimental data for improving their prediction capability. So, there is a trade-off between speed and accuracy, and the user has to decide his/her preference. A number of spectra prediction tools, commercial as well as open source, are discussed in this chapter accompanied with detailed tutorials on the use of some of them. To manage the data, many online servers and spectral databases are available today and a brief introduction to them is also provided. Here, we also describe an in-house-developed carbon and proton NMR chemical shift-based binary fingerprints and their use in virtual screening.