Machine learning–assisted structural characterization of ring size distribution in Sr-doped bioactive glasses: A molecular dynamics study

Bioactive glasses (BGs) are promising biomaterials in tissue engineering, orthopedics, and dentistry due to their ability to bond with living tissue. In this study, the atomic structure of strontium-substituted melt-quenched 60SiO2–(40–x)CaO–xSrO BGs (x = 0–20 mol%) was investigated using molecular dynamics (MD) simulations in LAMMPS. Structural data from radial distribution functions (RDFs) were used to train a two-dimensional convolutional neural network (2D CNN), which predicted ring size distribution (RSD) based on 10×100 matrices of atomic pair distributions. Short-range structural analysis showed a consistent Si–O bond length (1.607 Å) across samples, while Sr–O bond length was 2.57 Å with coordination numbers decreasing from 7.25 to 6.89. The Si–O–Si angle widened from 148.97° to 149.65° with rising Sr content, reflecting network expansion. Network-level parameters such as Qn unit distribution, bridging/non-bridging/free oxygens, and network connectivity (NC) were also evaluated. NC slightly decreased from 2.687 to 2.675, with minimal BO/NBO changes, and overall density rose due to the higher molar mass of Sr. Taken together, the study showed the effective integration of MD and machine learning framework, and the CNN accurately predicted RSD, achieving low MSE and high coefficient of determination R2.