Large Language Model Driven Predictive Modeling of Silver, Zinc, Titanium, Magnesium, Gold Doped Hydroxyapatite for Infrared Triggered Drug Delivery

Abstract In this paper, we report the application of Large Language Models (LLMs) in predictive modeling of infrared (IR)-triggered drug delivery systems, focusing on hydroxyapatite (HA) modified with silver nanoparticles (AgNPs), as well as dopants of zinc (Zn), titanium (Ti), magnesium (Mg), and gold (Au). We predict electronic structure, bandgap reduction, and optical properties utilizing real-world data integrating SMILES, benefiting from Darwin 1.5, a fine-tuned model, and T5, a predictive model, to predict electronic structure, bandgap values, and optical properties. Darwin 1.5 fine-tuned via question-answering and multi-tasking on scientific datasets correlates a mean absolute deviation (MAD) of 0.72 eV to bandgap predictions that are accurate and potentially cut the simulation time by as much as 50-70% compared to the conventional density functional theory (DFT) method. The T5 model enables simulations on optical properties via the computation of absorption spectra at a wavelength of 808 nm and the concentration-dependent roles on absorption and scattering. The predictions obtained from electronic structure calculations with a fine-tuned Darwin 1.5 model, combined with DFT analysis, indicate that Ag-HA's band gap reduction varied from 4.312 eV (0.25 mol% Ag doping) to 3.983 eV (0.75 mol% Ag doping); from about 4.68 eV (0.25 mol% Zn doping) to approximately 4.4 eV (0.75 mol% Zn doping) in praseodymium co-doped-HA; from around 3.8 eV (0.25 mol% Ti doping) to around 3.6 eV (0.75 mol% Ti doping); from about 4.61 eV (0.25 mol% Mg doping) to about 4.39 eV (0.75 mol% Mg doping); and from about 4.3 eV (0.25 mol% Au doping) to about 3.95 eV (0.75 mol% Au doping). Moreover, it was observed that the photothermal efficacies are higher, with a value of 18.8% (internal) to 0.11 L g −1 cm (external) under a concentration of 2% Ag-HA irradiated with a wavelength of 445nm, followed by comparable efficacies observed under similar conditions with a value of 11.6% (internal) to 0.055 L g −1 cm (external) with a concentration of Au-HA. Relatively low cell toxicity are observed in literature, in vitro studies show fairly balanced antibacterial activity, in addition to in vivo studies that reveal promoted bone healing with lowered systemic toxicity. The results are represented in detailed tables, figures, and Python codes. Ultimately, the present study serve to promote the application of LLMs in materials science, particularly in the area of cancer and infectious diseases, as well as considerations of the ethics of AI applications and the dangers of nanomaterials.