Machine Learning Analysis of Cytotoxicity Determinants in Nanoparticle-Based Rheumatoid Arthritis Therapies

Nanoparticle-based therapies have gained attention in
recent years
as promising treatments for rheumatoid arthritis (RA), due to the
potential offered for targeted delivery, controlled drug release,
and improved biocompatibility. A deep understanding of the factors
that drive cytotoxicity is crucial for safer and more effective nanomedicine
formulations. To systematically analyze the determinants of cytotoxicity
reported in the literature, we constructed a data set comprising 2,060
instances from 56 publications. Each instance was described by 23
features covering nanoparticle characteristics, cellular environment
factors, and assay conditions potentially associated with cytotoxicity.
Machine learning (ML) approaches were incorporated to gain deeper
insight into key cytotoxicity drivers. We combined Boruta for feature
selection, Random Forest (RF) for cytotoxicity prediction and feature
importance evaluation, and Association Rule Mining (ARM) for rule-based,
hidden pattern discovery. Boruta feature selection results identified
the drug and nanoparticle concentration, core–shell material,
and cell type as major determinants of cytotoxicity. The RF model
demonstrated a strong predictive performance, further confirming the
significance of these features. Moreover, ARM revealed high-confidence
association rules linking specific conditions, such as high drug concentrations
and poly(aspartic acid)-based systems, to cytotoxic outcomes. This
structured machine learning framework provides a foundation for optimizing
nanoparticle formulations that balance therapeutic efficacy with cellular
safety in RA therapy.