S. No. | Prediction/method | Limitations/disadvantages | Advantage/applications |
---|---|---|---|
1 | NMR spectra prediction | Based on machine learning Require huge data sets to be trained for effective findings Defining data requirement is difficult Probability of a poor model is high Training set should be unique Predictions are always accompanied by uncertainties [42] The predictor Chemaxon requires the uniqueness of chemical structures Creation of large, complex chemical structures is challenging Unable to predict higher frequency range compounds (60.0–1000 Hz) | Prediction is possible in one or two minutes even if the structure is huge and complicated Estimates the NMR signals of new and exciting compounds (60.0–1000 Hz) User-friendly and suitable for natural product dereplication More uniqueness for a new chemical in comparison to all others [273] Well-known method for facilitating structural elucidation Even for a molecule of polyatomic (H, O, C, S, N, B, F, I, Cl, Br, P, Si, As, Ge, Sn, and Se,), it can predict 13C and 1H NMR spectra [48] |
2 | UV–visible spectra prediction | The solvation model methanol produced delayed findings as compared to the solvation modal water The prediction of very complicated compounds takes significantly longer, around 5–12 h Some compounds provide the desired outcomes after 24 h The spectrum visualization gave positive findings, although the spectrogram saved was of poor quality | TD-DFT is one of the most common approaches to predicting UV–visible spectra Simple and not computationally too expensive [50] Alternative solvation models, investigators may compute the number of excited states of various substances The error is frequently lower than using the experimental technique The computation is completed considerably faster |
3 | Infrared and Raman spectra prediction | The optimization of the structure is extremely critical Little variations in the optimized structure lead to incorrect absorption frequency predictions Raman intensities are a more involved problem than IR intensities They depend much more on the experimental setup The prediction of very complicated compounds takes significantly longer, around 24–48 h Even so, the compounds do not always provide the desired outcomes Low quality of the spectrogram generated | The prediction was quite straightforward and user-friendly The functional code was not very complex Infrared/Raman predictors may be used to estimate the absorption frequencies of novel and exciting compounds For example, using the functional code “! Opt numFreq” to predict the IR and Raman spectra of a single molecule at the same time [52] |
4 | Mass spectroscopy predictions | The library matching strategy has a coverage issue Expensive and time-consuming for capturing additional spectra Prediction techniques are expensive to compute The performance of the forward prediction mode is weak | A variety of reference data sources, including the NIST, NIH, and EPA MS databases, are available for prediction Bidirectional prediction mode captures the fragmentation events more accurately Electron ionization can be accurately predicted using quantum chemistry calculations |
5 | Fluorescence spectroscopy predictions | Smaller data sets produce less accuracy Higher degree conjugated bonding patterns for emission spectra are required to be calculated Supportive chemometric techniques, such as PCA, are necessary | Lowering of the calculation’s computational cost It offers an easy method to perform effective geometry optimizations There is good agreement between the computed values of vertical emission energy |
6 | Electrochemistry predictions | First, as EIS data is inherently noisy, spectra regression and prediction at unknown frequencies are made more difficult Typically, tested frequencies have a predetermined number of points per decade that are log-spaced | Few deviations and a short experimentation time Active learning acquisition with GPs |