Machine Learning-Assisted Near-Infrared Spectral Fingerprinting for Macrophage Phenotyping.

Spectral fingerprinting has emerged as a powerful tool that is adept at identifying chemical compounds and deciphering complex interactions within cells and engineered nanomaterials. Using near-infrared (NIR) fluorescence spectral fingerprinting coupled with machine learning techniques, we uncover complex interactions between DNA-functionalized single-walled carbon nanotubes (DNA-SWCNTs) and live macrophage cells, enabling phenotype discrimination. Utilizing Raman microscopy, we showcase statistically higher DNA-SWCNT uptake and a significantly lower defect ratio in M1 macrophages compared to M2 and naive phenotypes. NIR fluorescence data also indicate that distinctive intraendosomal environments of these cell types give rise to significant differences in many optical features, such as emission peak intensities, center wavelengths, and peak intensity ratios. Such features serve as distinctive markers for identifying different macrophage phenotypes. We further use a support vector machine (SVM) model trained on SWCNT fluorescence data to identify M1 and M2 macrophages, achieving an impressive accuracy of >95%. Finally, we observe that the stability of DNA-SWCNT complexes, influenced by DNA sequence length, is a crucial consideration for applications, such as cell phenotyping or mapping intraendosomal microenvironments using AI techniques. Our findings suggest that shorter DNA-sequences like GT give rise to more improved model accuracy (>87%) due to increased active interactions of SWCNTs with biomolecules in the endosomal microenvironment. Implications of this research extend to the development of nanomaterial-based platforms for cellular identification, holding promise for potential applications in real time monitoring of cellular differentiation.