Machine learning guided Box-Behnken design optimized green synthesis of carbon nitride nanoparticles with antioxidant...

Nanomaterials (NMs) are gaining attention for their high surface area, tuneable reactivity, and potential in targeted therapeutic applications such as antibacterial, antifungal, and antioxidant effects. Their relevance is particularly noted in countering elevated reactive oxygen species (ROS), which drive oxidative stress (OS), DNA damage, inflammation, and diseases like cancer and neurodegeneration. In the commonly known neurodegenerative disorders such as Alzheimer's and Parkinson's disease, ROS accumulation leads to mitochondrial dysfunction, endoplasmic reticulum dysfunction neuronal death, and neuroinflammation. However, NM cytotoxicity remains a major concern. This study proposes a green, computationally guided synthesis of non-toxic antioxidant nanoparticles using Litsea glutinosa bark powder. A machine learning-based Nano Quantitative Structure-Activity Relationship (nano-QSAR) model was developed to identify structural features and screen for low-cytotoxicity nanoparticles using existing literature data. Gradient Boosting Regression achieved the highest predictive accuracy (R² = 0.85), guiding the selection of carbon-based nanoparticles due to their low cost and high biocompatibility. Carbon nitride nanoparticles from Litsea glutinosa (CNNP-LG) were synthesized using a hydrothermal method, with synthesis parameters- temperature, Litsea glutinosa powder, urea, and duration optimized via Box-Behnken Design. CNNP-LG was characterized using XRD, FTIR, and SEM-EDS techniques. Antioxidant activity assays showed an IC₅₀ of 15.62 µg/ml and 95% DPPH scavenging activity at 125 µg/ml, comparable to Ascorbic Acid (94.5%). Cytotoxicity analysis using the MTT assay confirmed high SH-SY5Y cell viability (87.8%) even at 500 µg/ml. To further understand the cytotoxic potential, the experimental results from CNNP-LG were incorporated into an existing carbon nitride nanoparticle dataset and used to validate a machine learning-guided nano-QSAR model developed for predicting cell viability. This integrated approach effectively reduces experimental workload, enhances nanoparticle safety profiling, and offers a promising pathway for developing safer nanomedicines with neurotherapeutic potential.