A Machine Learning Tool to Analyse Spectroscopic Changes in High-Dimensional Data

TL;DR

The paper addresses how nanoparticle surface chemistry drives temperature-dependent conformational changes of fibrinogen in the protein corona, using a multimodal spectral dataset (UVRR, CD, UV absorbance). It introduces a physics-informed, unsupervised ML workflow that combines the $W_1$ Wasserstein distance with t-SNE to quantify and visualize structural disorder across carbon and silica NP interfaces. The approach reveals that hydrophobic carbon NPs induce gradual, incomplete unfolding with limited aggregation, while hydrophilic SiO$_2$ NPs resemble free fibrinogen unfolding but with partial suppression, and the method robustly handles high-dimensional, multi-source data. These findings validate theoretical predictions and support design strategies for NP surfaces in nanomedicine, enabling quantitative analyses of protein–NP interactions across modal spectra.

Abstract

When nanoparticles (NPs) are introduced into a biological solution, layers of biomolecules form on their surface, creating a corona. Understanding how the structure of the protein evolves into the corona is essential for evaluating the safety and toxicity of nanotechnology. However, the influence of NP properties on protein conformation is not well understood. In this study, we propose a new method that addresses this issue by analyzing multi-component spectral data using Machine Learning (ML). We apply the method to fibrinogen, a crucial protein in human blood plasma, at physiological concentrations while interacting with hydrophobic carbon or hydrophilic silicon dioxide NPs, revealing striking differences in the temperature dependence of the protein structure between the two cases. Our unsupervised ML method a) does not suffer from the challenges associated with the curse of dimensionality, and b) simultaneously handles spectral data from various sources. The method offers a quantitative analysis of protein structural changes upon adsorption and enhances the understanding of the correlation between protein structure and NP interactions, which could support the development of nanomedical tools to treat various conditions.

Figures (9)

• Figure 1: 3D structure of Fib with atomic resolution determined by X-ray crystallography (PDB Entry: 3GHG). Different colors represent the various polypeptide chains of Fib: the $\alpha$ (blue), $\beta$ (orange), and $\gamma$ (pink) chains. It is a symmetric protein, with chains joined together at the central node, and terminal nodes for the $\beta$ and $\gamma$ chains. The aromatic amino acids are represented as green spheres and are mostly located in the terminal nodules and the central node, and much less in the alpha-helix. The dashed gray rectangle labeled "x2" highlights that the protein has two symmetric parts.
• Figure 2: Physicochemical characterization of the NPs and the Fib corona. (a) Transmission Electron Microscopy (TEM) image of CNPs with a nominal diameter of $\sim$120 nm. After image processing, their core diameter was calculated to be (118.3 $\ pm$40.8) nm. For pristine CNPs and CNPs in solution with Fib at WR between 0.1 and 1.0, as indicated in the legend, we measure the DLS distribution (b), consistently finding a polydispersity index (PDI) of less than 0.12, which is consistent with the DCS analysis (c). We repeat the same measurements for SiO$_2$NPs with a nominal diameter of 100 nm: image processing of TEM images estimates their diameter as (95.4 $\pm$ 7.8) nm (d); for pristine SiO$_2$NPs and SiO$_2$NPs in solution with Fib at WR between 0.1 and 1.0, as indicated in the legend, we always observe a PDI of 0.01 by DLS (e), consistent with the DCS analysis (f). Measurements were performed at ambient temperature.
• Figure 3: TEM image of Fib aggregates under dried conditions obtained after heating the sample of human Fib at blood-like concentration (2 mg/mL) up to 88$^\circ$C. The shape of the aggregate resembles the typical structure of fibrin fibers.
• Figure 4: UVRR spectra for free Fib and Fib with NPs. The 226 nm-excited UVRR spectra are collected for Fib at 2 mg/mL in PBS pH 7.4 (a), Fib with CNPs (b), and Fib with SiO$_2$NPs (c) as a function of temperature. These spectra are reported in two ranges: 1000-1800 and 2800-3800 $\mathrm{cm^{-1}}$. The main vibrational features in the Fib spectrum are labeled in panel (a). The temperature range and steps are indicated by the color scale on the right. Dashed lines in panels (b) and (c) show the spectra of Fib alone at the lowest temperature of 22$^\circ$C.
• Figure 5: Integrated area of some characteristic bands in the UVRR spectra in Fig. \ref{['fgr:spectra']}. (a) OH band area for Fib in PBS (black circles) and with CNPs (red pentagons), as well as SiO$_2$NPs (green squares). (b)-(e) Areas of the peaks in the frequency intervals reported in Table \ref{['tab:table1']} for Fib in PBS and with CNPs and SiO$_2$NPs. In each panel, the frequency range is indicated as a label, and the symbols are the same as in panel (a). Where indicated, the experimental data have been fitted with a sigmoidal curve (full black lines).