ATR-MIR spectral data and PLS outputs for protein measurement in reconstituted skim milk during passive warming
Description
This dataset accompanies the Data in Brief article "Dataset and reproducibility package for spectral-window ATR-MIR milk-protein predictions during passive warming of reconstituted skim milk" and supports the related *Measurement* article "Diagnosing temperature-induced bias and dispersion in ATR-MIR protein measurement using spectral-window PLS models". It examines sample temperature as a non-analyte measurement-state variable affecting ATR-MIR partial least-squares (PLS) protein predictions across spectral windows and evaluates an internal residual-versus-temperature correction. Eight independently prepared calibration formulations (S1W-S7W and S9W) were used to build and internally cross-validate seven one-latent-variable models: Full MIR, PhosCov, Amide I, Amide II, Amide III, Amide ALL and Amide I_II_III. One calibration spectrum was retained per formulation; a separate aliquot from the same prepared formulation was submitted for Kjeldahl analysis. Two additional liquids were separately prepared from the same skim-milk powder batch and to the same nominal 5.06% powder-to-water formulation as S2W. One was measured with three technical scans at 4.10 °C and three at 20.70 °C after 1 h equilibration at each condition. The other was measured sequentially during one passive-warming trajectory at nominal milestones of 2, 5, 8, 11, 14, 17 and 20 °C. Neither temperature-experiment sample was Kjeldahl analysed. The S2W laboratory value of 1.610% m/m protein is therefore supplied as a formulation-associated comparator from a separate calibration preparation, not as a contemporaneous reference measurement of either temperature-experiment sample. Existing output columns named "Benchmark" retain that legacy field name but have this comparator meaning. The repository provides formulation and laboratory metadata; complete 2018 liquid- and powder-sample reports; raw, processed and open-format ATR-MIR spectra; model and cross-validation outputs; warming and fixed-condition predictions; diagnostics; MATLAB scripts; figure-source data; supplementary material; checksums; and clean-run evidence. The fixed PhosCov objects retain legacy TS0486 labels and are a separate spectral-window acquisition sequence from the same fixed preparation as the TS0506 Full MIR/Amide data; the original labels and values are preserved. Within this controlled source-data envelope, PhosCov has the lowest RMSECV and smallest warming-trajectory prediction dispersion, Full MIR has the smallest root-mean-square deviation from the formulation-associated comparator, and Amide I/II show the largest temperature-associated prediction changes. Leave-one-temperature-out correction reduces PhosCov root-mean-square and mean signed comparator deviation but not dispersion. These descriptive results do not constitute external validation across independent batches, natural milk matrices, instruments or process environments.
Files
Steps to reproduce
Download and extract Measurement_Paper1_Repository_CORRECTED.zip. In MATLAB R2024b, ensure Eigenvector Research PLS_Toolbox 9.5.1 is on the MATLAB path. Open the extracted 06_MATLAB_Code folder and run: clear; close all force; clc result = MEASUREMENT_RUN_REPOSITORY; The runner validates the controlled inputs, rebuilds the seven one-latent-variable PLS models, regenerates predictions, tables, diagnostic data, spectral exports and figures, and writes 71 generated artifacts plus a SHA-256 manifest to a new timestamped temporary folder without modifying the deposited files. Open-format calibration, passive-warming and fixed-condition spectra are also supplied as CSV files for inspection without MATLAB. Full requirements and optional output-folder syntax are provided in 06_MATLAB_Code/README_PUBLIC_RUNNER.md.
Institutions
- University of Southern QueenslandQueensland, Toowoomba