A new analytical approach combining terahertz spectroscopy with machine learning could support faster, non-destructive monitoring of active pharmaceutical ingredient content.
Researchers have combined terahertz (THz) spectroscopy with chemometric and machine-learning algorithms to identify and quantify active pharmaceutical ingredients (APIs) in multi-component pharmaceutical samples.
The study, published in Analytical Methods, reported predictive accuracy of up to R² = 0.999 and reductions in prediction time of as much as 90 percent.
Xiaoning Wang et al. used principal component analysis (PCA) alongside machine-learning models to interpret THz spectral data. The researchers reported that the resulting approach could perform both qualitative identification and quantitative content analysis, with robustness assessed using pharmaceutical systems containing multiple components.
THz spectroscopy measures the interaction between electromagnetic radiation and a sample in the terahertz frequency range. Because different molecular structures and crystalline forms can produce characteristic spectral responses, the technology has potential for distinguishing APIs from excipients without destroying the sample.
Viewed through a process analytical technology (PAT) lens, the combination could offer a route towards rapid at-line or potentially in-line measurement of API concentration. Faster analysis could allow manufacturers to identify variability during production instead of waiting for results from conventional laboratory testing.
the combination could offer a route towards rapid at-line or potentially in-line measurement of API concentration”
The use of machine learning is significant because pharmaceutical spectra can contain overlapping signals arising from the API, excipients and physical properties of the sample. Algorithms trained to recognise these complex relationships may extract quantitative information that is difficult to obtain through conventional spectral interpretation.
However, the reported performance represents an experimental proof of concept rather than a validated pharmaceutical manufacturing method. GMP adoption would require testing across independent batches, instruments, manufacturing sites and expected sources of process variation. Manufacturers would also need to demonstrate model lifecycle control, data integrity and the continued suitability of calibration datasets.
Changes to raw materials, formulations or processing conditions could affect spectral responses and therefore model performance. These risks would need to be addressed through prospective validation and ongoing model monitoring.
Nevertheless, the findings indicate that pairing rapid, non-destructive spectroscopy with machine learning could extend the capabilities of pharmaceutical PAT. If translated successfully into manufacturing, the approach could support earlier process interventions, improved content uniformity and more data-rich control strategies.
No comments yet