Machine-Learning-Based Approach to Decode the Influence of Nanomaterial Properties on Their Interaction with Cells.

In an in vitro nanotoxicity system, cell-nanoparticle (NP) interaction leads to the surface adsorption, uptake, and changes into nuclei/cell phenotype and chemistry, as an indicator of oxidative stress, genotoxicity, and carcinogenicity. Different types of nanomaterials and their chemical composition or "corona" have been widely studied in context with nanotoxicology. However, rare reports are available, which delineate the details of the cell shape index (CSI) and nuclear area factors (NAFs) as a descriptor of the type of nanomaterials. In this paper, we propose a machine-learning-based graph modeling and correlation-establishing approach using tight junction protein ZO-1-mediated alteration in the cell/nuclei phenotype to quantify and propose it as indices of cell-NP interactions. We believe that the phenotypic variation (CSI and NAF) in the epithelial cell is governed by the physicochemical descriptors (e.g., shape, size, zeta potential, concentration, diffusion coefficients, polydispersity, and so on) of the different classes of nanomaterials, which critically determines the intracellular uptake or cell membrane interactions when exposed to the epithelial cells at sub-lethal concentrations. The intrinsic and extrinsic physicochemical properties of the representative nanomaterials (NMs) were measured using optical (dynamic light scattering, NP tracking analysis) methods to create a set of nanodescriptors contributing to cell-NM interactions via phenotype adjustments. We used correlation function as a machine-learning algorithm to successfully predict cell and nuclei shapes and polarity functions as phenotypic markers for five different classes of nanomaterials studied herein this report. The CSI and NAF as nanodescriptors can be used as intuitive cell phenotypic parameters to define the safety of nanomaterials extensively used in consumer products and nanomedicine.