NIR Model Transferability Using Binary Mixtures of Talc in Iron Sulfate and Water in Ethanol

Importance of the Topic

Near-infrared (NIR) spectroscopy is widely adopted for rapid, non-destructive quantitative analysis across pharmaceutical, chemical and polymer industries. A key barrier to routine deployment at scale is calibration transferability — the ability to apply a model built on one instrument (primary) directly to other instruments (vector) without extensive mathematical correction. Demonstrating robust transferability reduces cost, time and regulatory burden associated with redeveloping calibrations and supports decentralized quality control and process monitoring.

Objectives and Study Overview

This study evaluates straightforward transferability of quantitative FT-NIR calibration models between multiple Thermo Scientific Antaris instruments using two model systems representing different spectral challenges: talc dispersed in an inert (iron sulfate) matrix (solid, low-intensity, sharp non-shifting bands) and binary liquid mixtures of ethanol and water (strong absorbers with concentration-dependent band shifts due to hydrogen bonding). The goals were to assess whether calibrations developed on a primary Antaris could predict identical samples measured on other Antaris instruments without post-transfer corrections, and to identify instrumental and chemometric factors enabling successful transfer.

Experimental Design and Methodology

• Sample systems: (1) Talc in ammonium iron(III) sulfate dodecahydrate — dry-mixed solid samples covering low to several percent talc. (2) Ethanol/water binary mixtures prepared gravimetrically across the full range (0.1% to 100% water) with independent sample preparation (no serial dilutions).
• Instrument matching assessment: A toluene subtraction test was used to inspect x-axis (wavelength) and general optical sameness between instruments prior to transfer attempts. Polystyrene is also noted as an alternative external standard.
• Chemometric approach: Calibrations were developed on a primary instrument using Partial Least Squares (PLS) and Stepwise Multiple Linear Regression (SMLR). Pre-treatments included Multiplicative Scatter Correction (MSC), Norris derivatives, and Savitzky–Golay derivatization for offset correction. Models were applied to vector-instrument spectra without slope/bias adjustments or algorithmic transfer corrections to test direct transfer.

Used Instrumentation

• Antaris FT-NIR analyzers (multiple units): internal reference laser for improved x-axis stability (Conne’s Advantage) and InGaAs detectors.
• Talc/iron sulfate (diffuse reflectance) configuration: integrating sphere, diffuse reflectance mode; spectral range 3800–12,000 cm01; resolution 4 cm01; 90 co-averaged scans; Norton-Beer medium apodization; ~67 s acquisition.
• Ethanol/water (transmission) configuration: liquid transmission module with 0.5 mm quartz cuvette; spectral range 3800–12,000 cm01; resolution 8 cm01; 64 co-averaged scans; ~32 s acquisition; Norton-Beer medium apodization; InGaAs detector.
• Reference materials and reagents: high-purity toluene for matching test, distilled water, ethanol (KF-checked), talc and ammonium iron(III) sulfate.

Main Results and Discussion

• Toluene matching: Toluene subtraction spectra indicated strong instrument matching for the instrument pairs used in both talc and ethanol/water studies, providing confidence that hardware differences would not dominate transfer outcomes.
• Talc (solid model): Talc exhibits narrow, sharp bands (e.g., the 7185 cm01 band with half-height width ~6.1 cm01) and relatively small photometric response. A calibration developed on one Antaris was used to predict identical samples measured on five other Antaris units. Predictions across six instruments showed small absolute differences (mostly in the second decimal place in percent talc) and low relative standard deviations (%RSD typically a few percent and decreasing with higher talc concentration). Correlation slopes and intercepts for vector instruments fell within the 95% confidence limits of the primary-instrument calibration parameters, indicating excellent transfer performance despite the narrow-band sensitivity to x-axis accuracy.
• Ethanol/water (liquid model): This system is spectrally more complex because hydrogen-bonding produces concentration-dependent band shape and frequency shifts in both combination and overtone regions. Despite this nonlinearity, a 2-factor PLS model built on the primary instrument (using 10–90% ethanol) produced nearly identical predictions when applied to spectra from a vector instrument. Diagnostic overlays (second-derivative spectra for selected concentrations), percent intensity differences at key frequencies (generally <1% relative, except at the lowest level 0.1% water where relative differences were ~4%), and principal component score overlays (first two PCs explaining ~90% of spectral variance) all demonstrated that the two instruments produced essentially the same spectral data and that direct transfer was possible.

Benefits and Practical Implications

• FT-NIR analyzers with well-controlled optical components and an internal laser reference provide robust wavelength stability that aids cross-instrument transfer without additional mathematical matching.
• Successful direct transfer minimizes the need for empirical adjustments (slope/bias), spectral matching algorithms, or the time-consuming process of inoculation (adding vector-instrument spectra to the calibration) or full redevelopment on each instrument — offering substantial savings in validation effort for regulated environments.
• Routine instrument assessment using simple external tests (toluene or polystyrene subtraction) can provide early indication of instrument readiness for transfer and guide troubleshooting when transfer fails.

Future Trends and Potential Applications

• Standardized readiness tests and transferability protocols are likely to be adopted more widely, enabling scalable deployment of NIR across multiple sites and instruments in manufacturing networks.
• Advances in chemometrics and machine learning could yield transfer-invariant features or models that further reduce the need for instrument-specific corrections or inoculation, supporting centralized calibration libraries and cloud-based model distribution.
• Automated instrument qualification (including routine toluene/polystyrene checks) and enhanced factory control of optical components will improve baseline sameness and reduce variability across instrument fleets.
• Hybrid approaches combining physical matching, minimal empirical correction (slope/bias), and selective inclusion of representative vector-instrument spectra will remain practical strategies where perfect instrument sameness cannot be guaranteed.

Conclusion

This study demonstrates that, with well-matched Antaris FT-NIR instruments and careful data acquisition, direct transfer of quantitative calibrations for both a challenging solid model (talc in iron sulfate) and a nonlinear liquid model (ethanol/water) is feasible without additional transfer corrections. Key enablers were instrument wavelength and photometric consistency (verified by toluene tests), appropriate spectral pre-treatment, and robust chemometric modeling. These model systems may serve as routine checks for instrument readiness and support broader implementation of directly transferable NIR methods in quality control and process analytics.

References

1. J.K. Drennen, E.G. Kramer and R.A. Lodder, Critical Reviews in Analytical Chemistry, 22(6), 443 (1991).
2. J.D. Kirsch and J.K. Drennen, Applied Spectroscopy Reviews, 30(3), 139 (1995).
3. K.M. Morisseau and C.T. Rhodes, Drug Development and Industrial Pharmacy, 21, 1071 (1995).
4. K.H. Norris, Journal of Near Infrared Spectroscopy, 4, 69 (1996).
5. E.W. Ciurczak, Pharmaceutical Technology, 15, 140 (1991).
6. W.F. McClure, Analytical Chemistry, 66, 43A (1994).
7. J. Lin, Applied Spectroscopy, 52, 1591 (1998).
8. J.S. Shenk, M.O. Westerhaus and W.C. Templeton Jr., Crop Science, 25, 159 (1985).
9. J.S. Shenk and M.O. Westerhaus, Crop Science, 31, 1694 (1991).
10. Y. Wang and B.R. Kowalski, Applied Spectroscopy, 46, 764 (1992).
11. Y. Wang and B.R. Kowalski, Analytical Chemistry, 65, 1301 (1993).
12. Y. Wang, M.J. Lysaght and B.R. Kowalski, Analytical Chemistry, 64, 562 (1992).
13. Z. Wang, T. Dean and B.R. Kowalski, Analytical Chemistry, 67, 2379 (1995).
14. E. Bouveresse, D.L. Massart and P. Dardenne, Analytica Chimica Acta, 297, 405 (1994).
15. E. Bouveresse, D.L. Massart and P. Dardenne, Analytical Chemistry, 67, 1381 (1995).
16. E. Bouveresse, D.L. Massart, I.R. Last and K.A. Prebble, Analytical Chemistry, 68, 982 (1996).
17. T.B. Blank, S.T. Sum, S.D. Brown and S.L. Monfre, Analytical Chemistry, 68, 2987 (1996).
18. J. Lin, S.-C. Lo and C.W. Brown, Analytica Chimica Acta, 349, 263 (1997).
19. F. Despagne, B. Walczak and D.L. Massart, Applied Spectroscopy, 52(5), 732 (1998).
20. J. Workman and J. Coates, Spectroscopy, 8(9), 36 (1993).
21. S.R. Lowry, J. Hyatt and W.J. McCarthy, Applied Spectroscopy, 54(3), 450 (2000).