ATR-FTIR combined with PCA and PLS1 models as a screening tool: performance in single-adulterant systems and limitations under multi-component conditions in UHT milk

Abstract The simultaneous adulteration of UHT milk with multiple compounds presents analytical challenges due to spectral overlap, matrix effects, and collinearity. This study evaluated ATR-FTIR spectroscopy combined with PCA and analyte-specific PLS1 models, aiming to screen for and quantify seven adulterants and empirically assess PLS1 model behavior under multicomponent interference. PCA revealed distinct score-space patterns among the evaluated groups, although the observed separation could also include brand-related matrix effects. For independent samples containing a single adulterant, PLS1 models demonstrated satisfactory predictive performance for sodium hypochlorite, sodium bicarbonate, formaldehyde, urea, and sodium hydroxide (R 2 P > 0.91; RPD ≥ 3.0), whereas starch and hydrogen peroxide showed inferior predictive capacity (R 2 P = 0.71–0.85; RPD = 2.00–2.79). Under multicomponent conditions, however, model selectivity decreased markedly, producing analyte-dependent false-positive and false-negative responses in the symmetric scenario, while predictions in the asymmetric scenario were sporadic and unstable. These results indicate that ATR-FTIR combined with PCA holds promise for exploratory screening and that PLS1 enables the prediction of specific analytes within defined calibration domains. Nevertheless, single-adulterant PLS1 models are unreliable for quantifying adulterants under the multicomponent scenarios evaluated.