Advanced Near IR Algorithm Compensates for Spectral Features Related to Changes in Sampling Vials

Significance of the topic

Fourier transform near-infrared (FT-NIR) spectroscopy is widely used for rapid, non-destructive quantitative analysis of liquids in chemical and pharmaceutical settings. Low-cost disposable sampling vials reduce analysis time and contamination risk, but variability in vial materials or suppliers can introduce subtle spectral features that degrade the performance of calibration models. Addressing spectral variance arising from sampling hardware is therefore essential for robust method transfer, routine QA/QC, and maintaining traceable quantitative results across instruments and sites.

Objectives and study overview

This technical note evaluates how differences between disposable culture tubes from two vendors affect FT-NIR spectra and demonstrates an augmented classical least squares (ACLS) approach to compensate for vial-related spectral variance. The study compares classical least squares (CLS) and ACLS calibrations for the determination of water content in ethanol, showing how transfer standards and augmentation shapes can restore predictive accuracy when sampling vials change.

Methodology

Samples: Binary mixtures of water in ethanol (0–10% water) were prepared to serve as calibration and validation standards. Two sets of standards were measured: one in Fisher Scientific disposable culture tubes and one in Kimble Kontes tubes.

Spectral acquisition: Transmission FT-NIR spectra were collected on a Thermo Scientific Antaris FT-NIR analyzer at 8 cm^-1 spectral resolution with 0.5 minute measurement time per sample. No temperature control was applied to the samples. Spectra from empty tubes were also acquired to characterize baseline differences between suppliers.

Calibration strategies: Four calibration scenarios were constructed to compare CLS and ACLS performance:

1. Calibration 1 — CLS: Method and validation standards both measured in Fisher tubes (baseline performance).
2. Calibration 2 — CLS: Calibration in Fisher tubes, validation in Kimble tubes (tests sensitivity to change in vial supplier).
3. Calibration 3 — ACLS without transfer standards: method standards in Fisher tubes, Kimble spectra placed on a transfer tab but not used for calibration (compare ACLS vs CLS without explicit transfer information).
4. Calibration 4 — ACLS with transfer standards: method standards in Fisher tubes and three Kimble spectra used as calibration transfer standards; remaining Kimble spectra used for validation (tests effect of adding transfer shapes).

Used instrumentation

The primary instrument used was a Thermo Scientific Antaris FT-NIR spectrometer configured for liquid transmission measurements. Data acquisition parameters: 8 cm^-1 resolution, 0.5 minute per sample. The study used standard laboratory disposable culture tubes from two vendors: Fisher Scientific and Kimble Kontes.

Theory and ACLS algorithm summary

Classical least squares (CLS) models represent mixture spectra as linear combinations of pure-component spectra; CLS assumes that all spectral variance in unknowns is described by these components. When spectral variance unrelated to analytes (for example, vial material features) is present, CLS predictions can be biased.

The ACLS approach retains the interpretability of CLS while adding orthogonal shapes (vectors) to model unexplained spectral variance. Two sets of shapes are generated: (1) shapes derived from residuals of the method standards (to model variability present during initial calibration) and (2) shapes derived from residuals of transfer standards (to capture additional variance present in transferred samples). The method builds augmented calibration matrices (Ks and Kt) and computes unknown concentrations using a generalized inverse formulation: Cunk = [Kt^T Kt]^-1 Kt^T Aunk, where Aunk is the spectrum of the unknown and Kt contains component and augmentation shapes. The optimal number of augmentation shapes is determined by minimizing cross-validation error.

Results and discussion

Spectral comparison: Average spectra of empty Fisher and Kimble tubes differ subtly (peak magnitude <0.002 absorbance units), with a notable feature near 7000 cm^-1 — a region relevant to water absorption. These small differences were sufficient to degrade prediction accuracy when the CLS model calibrated on Fisher tubes was applied to samples measured in Kimble tubes.

Calibration performance (summary):

• Calibration 1 (CLS, Fisher-only): RMSEC = 0.0448% water; RMSEP = 0.0637% water.
• Calibration 2 (CLS, calibration Fisher; validation Kimble): RMSEC = 0.0487% water; RMSEP = 0.242% water — approximately fourfold degradation in prediction error caused by change in tube supplier.
• Calibration 3 (ACLS, no transfer standards): RMSEC = 0.0373% water; RMSEP = 0.256% water — ACLS without transfer spectra did not repair prediction error.
• Calibration 4 (ACLS with 3 transfer standards from Kimble): RMSEC = 0.0841% water; RMSEP = 0.0291% water — adding two transfer-derived shapes increased calibration error but greatly improved prediction on Kimble validation spectra.

The key outcome is that including a small number of transfer standards (three spectra) and allowing ACLS to derive two augmentation shapes captured the vial-related spectral variance and restored prediction accuracy to better than the original CLS baseline. The increase in RMSEC when adding transfer spectra reflects the broader variability introduced into the calibration set, but RMSEP (the metric relevant to external predictions) was substantially reduced.

Figures and table content (qualitative): Figures show the Antaris instrument setup, calibration diagnostics, and mean spectra comparisons; Table 1 summarizes RMSEC and RMSEP for the four calibration scenarios as listed above. The mean subtraction spectrum highlights the vial-specific spectral feature near 7000 cm^-1.

Benefits and practical applications

• Robust method transfer: ACLS enables retention of the original CLS model while augmenting it to accommodate new, instrument- or consumable-related spectral variance, facilitating calibration transfer among instruments and sites.
• Minimal additional standards: A small number of transfer standards (often just a few spectra) can be sufficient to model new variance sources without rebuilding the entire calibration.
• Interpretability: Because ACLS is built on CLS, spectral contributions from defined chemical components remain explicit, which supports forensic interpretation and regulatory traceability in QA/QC environments.
• Implementation: The approach is implemented in Thermo Scientific TQ Analyst software, where transfer shapes are stored in a separate library leaving the original method unchanged.

Future trends and potential applications

• Broader deployment for method transfer across multi-site manufacturing and contract laboratories where consumable variability is common.
• Automation of optimal transfer standard selection using active learning or leverage diagnostics to minimize the number of required transfer spectra.
• Integration with instrument performance qualification (IQ/OQ) workflows to automatically detect spectral deviations attributable to accessories or sample holders and apply ACLS corrections.
• Extension to other spectroscopic modalities (mid-IR, Raman) and to solid sample holders where substrate variability affects spectra.

Conclusion

This study demonstrates that subtle spectral differences introduced by disposable sampling vials can substantially degrade CLS model predictions in FT-NIR analyses. The ACLS algorithm, by augmenting a CLS model with shapes derived from method and transfer residuals, effectively compensates for vial-related variance. Using only a few transfer standards to generate augmentation shapes restored and even improved prediction performance on samples in different tubes, while preserving the interpretability of the original CLS model. ACLS is therefore a practical and efficient strategy for robust method transfer and maintaining quantitative accuracy when sampling conditions change.

References

• Lowry S., McCarthy W., Ritter G., Thermo Fisher Scientific. Advanced Near IR Algorithm Compensates for Spectral Features Related to Changes in Sampling Vials. Technical Note 51696.
• Haaland D.M., Melgaard D.K., Sandia National Laboratories. (Described in text as the source of the augmented CLS concept.)