Two-dimensional correlation FTIR spectroscopy combined with machine learning reveal cadmium induced biomolecular alterations in Clarias batrachus

This study applies two-dimensional correlation spectroscopy (2DCOS) and perturbation-correlation moving window (PCMW) 2DCOS to examine molecular alterations caused by cadmium (Cd) toxicity in Clarias batrachus fingerlings. Synchronous spectra revealed auto peaks near ∼3048 and 2970 cm−1, linked to unsaturated fatty acids and methyl lipid groups. Asynchronous maps indicated variations in CH2 and CH3 fatty acids, suggesting altered membrane structure. Protein modifications were also observed, with transitions from α-helices to β-turns reflecting instability. PCMW2DCOS highlighted Cd concentration-dependent effects: 0.8–1.9 ppm primarily disrupted lipid structures, while lower doses (0.4–1.2 ppm) signaled early protein changes. At higher levels (3.2–4.5 ppm), carbohydrate alterations became evident. These results emphasize the potential of PCMW2DCOS to detect subtle biomolecular responses to heavy metal stress on edible aquatic species. Additionally, Fourier-transform infrared (FTIR) spectroscopy coupled with machine learning was used to classify muscle tissues under Cd exposure. The PCA-LDA model achieved 96% accuracy with robust metrics, including an F score of 96% and MCC of 0.94, though minor misclassifications appeared in the confusion matrix. Support Vector Machine (SVM) models also performed strongly, with linear and polynomial kernels achieving 96% accuracy, while RBF and Sigmoid kernels were slightly lower at 91% and 87%. In conclusion, the integration of 2DCOS with FTIR and machine learning improved the detection of early biomolecular disruptions that are not visible in conventional one-dimensional FTIR. This approach provides new insight into Cd toxicity in edible fish species highlighting their utility in environmental toxicology and food safety monitoring. HIGHLIGHTS2DCOS spectra in the amide I region indicate a transition from α-helices to β-turns, suggesting protein instability due to Cd exposure.PCMW2DCOS asynchronous spectral changes suggest phase-wise alterations in glycogen metabolismSVM models with linear and polynomial kernels demonstrated the highest classification performance, achieving an accuracy of 97%.