Table 10 Chromatography behaviour predictions’ limitations and benefits
S. No | Prediction/method | Limitations/disadvantages | Advantage/applications |
|---|---|---|---|
1 | QSRR and software | Prediction must require more than 50 compounds for study Statistical models that extrapolate outside the training set’s retention times may produce erroneous findings Even if the training set included a range of chemically varied molecules with structures that differ from the training sets are unlikely to be successfully predicted To accommodate the changes in the more complicated compounds, the training set must be bigger With an equal-sized training set, lipids provide better predictions than metabolites It is insufficient to separate very near co-elute compounds It is difficult to predict retention time when using several columns and conditions | The predictor places the majority of predictions within one or two minutes/ One minute of their real value equals about 5–8% of the retention time Approximately, all predictions fall between 2–5 min about 10–25 per cent depending on the column It will quickly and easily produce and save a large number of models In-house data tests yielded comparable results To estimate retention times for numerous columns and conditions It is adequate to enhance confidence in exact identifications Compound of the same interests that are clearly identified |
2 | AQbD | When overcome, can result in method failures and, in certain cases method replacements Chromatographic methods where the number of analytes is required for effective separation Since there are so many factors that impact the method’s outcomes Applying the AQbD paradigm to analytical methods is justifiable Instrument settings, sample characteristics, procedure parameters, and calibration model selection are examples of these factors [193] | Enhanced method efficiency; fewer trials, resulting in lower method cost; time utilization; levels of compliance; and knowledge of the extremes As the technique demonstrates a link between the method variables and performance The analyst gains confidence in the method’s effectiveness Analytical techniques are re-evaluated regularly to resolve any gaps in method performance To avoid failures in method transfer, OOS and OOT, AQbD methodology might be used AQbD allows for regulatory flexibility, but it necessitates to high level of robustness |
3 | Chemometrics | Some good laboratory-based analysers are not all mathematically minded, so they did not want to overburden their studies with maths Teaching and learning, chemometrics is still having problems integrating itself Chemometrics is partial because the fundamental body of information is overburdened Any new content must supplant previous subjects Basic statistical knowledge, such as univariate calibration, precision, accuracy, and uncertainty, is necessary [275] | Chemometrics technique advancement is continual, rapid, and efficient With the improvements in exploratory tools, they provide rich information about chemical systems [276] Adaptability for analysis of complex chemical process data in the industry [277] Quality control of herbal drugs, food analysis like vegetables [278,279,280], fruits [281] grains [282, 283], proteins [284] etc. Environmental chemistry studies [285] and assessment of the results The development of high-throughput chromatographic and spectroscopic data calculations |
3.1 | ANN | Retention time of new congener cannot be predicted No information about the relationship between molecular property and retention behaviour can be obtained [286] | An accurate and reliable Human brain way it is working Optimize the separation without employing analyte properties |
3.2 | SVM | Introducing bias in the results Class-modelling technique should only consider sensitivity when deciding on the ideal circumstances for process parameters | Exhibit better overall performance These models and the experimental result agree well The distribution of non-target samples is prevented from the target samples |
3.3 | MLR-ALS | Fail to develop an appropriate QSPR model Less prediction when compare with the SVM model [287] | Measuring the amount of variables Improved selectivity through improved chemical information separation from interference effects and higher signal-to-noise ratios, which improve chemical distribution visualization |
3.4 | PARAFAC | In some cases, PARAFAC is too restrictive to model the data in a sensible way Problems with scattering and missing values, for instance, have attracted specific focus | Quicker and more reliable computationally than SVD Applications include five-way data analysis and analysis of variance Relatively straightforward mathematically |
3.5 | PCR and PLSR | Risk of overfitting due to bias-variance trade-off Differences between bias and variance can have a significant impact on predictions [288] | Multi-resolution of multi-component mixtures |
3.6 | PLS | The method has more challenging Minor inconvenience compared to the risk of change correlation [291] Overstate the baseline when additive noise is present | A six-component combination was resolved using PLS [292] It is used to build multivariate calibration models [293] |
3.7 | PCA | Sample load at the small scale was insufficient to define the second principal compound Without a mathematical model the analysis is inaccurate Direct PCA is not suitable for the raw data | Predict chromatogram shapes at different scale of operation [294] Single principal compound is sufficient to explain most |
3.8 | TLRC and MLRC | Cannot be employed more than three compound mixtures Inaccurate calculations are caused by errors in wavelength set selection | Commonly employed in ternary mixture multi-resolution [164] A combination of strongly overlapping spectra, the accuracy is in the region of 99–101% More reliable with simple mathematical calculation [295] |