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

Importance of the topic

The accuracy and robustness of quantitative FT-NIR methods depend not only on the chemical calibration standards but also on sampling hardware and other sources of spectral variance. Low-cost disposable culture tubes are widely used for rapid liquid analysis, but small spectral differences between tube vendors can degrade predictions when classical least squares (CLS) calibrations are transferred to samples measured in different containers. The augmented classical least squares (ACLS) approach provides a practical, interpretable way to compensate for such variance and to support reliable method transfer across sampling conditions and instruments.

Objectives and study overview

This technical note evaluates an ACLS implementation (in Thermo Scientific TQ Analyst) for compensating vial-related spectral features that are not present in the original calibration set. The case study measures water concentration in ethanol (0–10% water) and compares CLS and ACLS performance when samples are measured in disposable culture tubes from two different suppliers (Fisher Scientific and Kimble Kontes). The goal is to assess whether a small number of transfer standards and added spectral shapes can restore prediction accuracy for samples measured in the alternate tubes.

Used instrumentation

• Thermo Scientific Antaris FT‑NIR analyzer, transmission module.
• Spectral acquisition parameters: 8 cm−1 resolution, ~0.5 minutes measurement time per sample.
• Disposable culture tubes from two vendors: Fisher Scientific (original calibration) and Kimble Kontes (transfer/validation).

Methodology

A series of ethanol–water mixtures (up to 10% water) were prepared as calibration and validation standards. No temperature control was applied during measurements to reflect typical routine conditions. Four calibration scenarios were evaluated:

1. CLS: calibration and validation standards all measured in Fisher tubes (baseline).
2. CLS: calibration in Fisher tubes, validation in Kimble tubes (demonstrate transfer failure).
3. ACLS without transfer standards: method standards in Fisher tubes; Kimble spectra assigned as validation (assess ACLS without explicit transfer information).
4. ACLS with transfer standards: method standards in Fisher tubes plus three Kimble spectra entered as calibration transfer standards; remaining Kimble spectra used for validation.

ACLS theory (as implemented) uses a two‑stage approach: build a CLS calibration from method standards, compute residual spectra Ea = A − KC, derive additional spectral shape vectors from the residuals of the method standards (Ks), and then derive transfer shapes from residuals of transfer standards (Kt). The final prediction uses an augmented Kt matrix and the equation Cunk = [KtT Kt]−1 KtT Aunk. The optimal number of additional shapes is selected by cross‑validation.

Main results and discussion

Spectral comparison of empty tubes from the two vendors shows subtle but measurable differences (absorbance differences < 0.002), including a feature near 7000 cm−1 within the water calibration region. These small differences substantially affected CLS predictions when validation spectra were acquired in Kimble tubes instead of Fisher tubes.

Key calibration metrics (RMSEC and RMSEP) for the four scenarios were:

• CLS (Fisher calib/validation): RMSEC ≈ 0.0448% water; RMSEP ≈ 0.0637% water.
• CLS (Fisher calib / Kimble validation): RMSEC ≈ 0.0487% water; RMSEP ≈ 0.242% water (fourfold increase in prediction error).
• ACLS (no transfer standards): RMSEC ≈ 0.0373% water; RMSEP ≈ 0.256% water (no improvement in validation error).
• ACLS (with 3 Kimble transfer spectra): RMSEC ≈ 0.0841% water; RMSEP ≈ 0.0291% water (significant improvement in prediction error).

The ACLS model that included three transfer standards introduced two additional spectral shapes derived from the transfer residuals. Although RMSEC rose (reflecting a larger combined calibration set and added variance), RMSEP for the Kimble validation set dropped substantially to ~0.029% water, outperforming the original CLS model and demonstrating effective compensation for vial‑related variance. Importantly, ACLS preserves the CLS interpretability of component contributions while allowing targeted augmentation for extraneous spectral effects.

Benefits and practical applications

• Robust method transfer: ACLS enables existing CLS calibrations to be extended to new sampling conditions (different vials, cell windows, or instruments) with a small number of transfer standards rather than full recalibration.
• Interpretability: retains CLS basis spectra for clear cause‑and‑effect interpretation while isolating non‑analyte variance in augmenting shapes.
• Operational efficiency: fewer transfer standards and separate storage of transfer shapes streamline routine updates and reduce downtime.
• Regulatory and QA/QC friendliness: because the primary CLS model remains unchanged, ACLS facilitates traceable method updates and documentation.

Future trends and potential applications

• Standardized transfer libraries: creation and sharing of transfer‑shape libraries for common sampling formats and instrument families to simplify method portability.
• Automated selection of transfer spectra: machine‑assisted identification of minimal representative transfer standards to reduce operator effort.
• Hybrid modeling: combining ACLS with robust preprocessing and PLS/PCR diagnostics to balance interpretability and predictive power when multiple unrelated variance sources exist.
• Expanded use cases: adapting ACLS for solid sampling accessories, flow cells, varying pathlengths, and multi‑site instrument networks.
• Regulatory adoption: defining best practices for using ACLS in validated workflows, including guidelines for the number and variety of transfer standards and cross‑validation procedures.

Conclusion

The ACLS approach demonstrated here effectively compensates for small but impactful spectral differences introduced by disposable sampling vials. By augmenting a CLS calibration with a small number of transfer‑derived spectral shapes, prediction accuracy for samples measured under the new sampling condition improved markedly with minimal additional experimental burden. ACLS preserves the interpretability of CLS while providing a pragmatic solution for method transfer and robustness in routine FT‑NIR quantitative analysis.

References

• Lowry S., McCarthy W., Ritter G., Advanced Near IR Algorithm Compensates for Spectral Features Related to Changes in Sampling Vials. Thermo Fisher Scientific Technical Note 51696, 2008.
• Haaland D. M., Melgaard D. K., Sandia National Laboratories. Foundational work on CLS augmentation and related multivariate calibration strategies (as referenced in the technical note).