Computational and regression analysis of neuroprotective agents using fuzzy neural networks and topological descriptors

Traumatic brain injury is still a serious neurological disorder in need of the creation of useful therapeutic agents with favorable physicochemical profiles. In this work, a computational approach is used that incorporates graph theoretical molecular descriptors, particularly topological indices into supervised machine learning models to forecast the physicochemical properties of potential drugs for traumatic brain injury therapy. A wide range of topological descriptors were retrieved from molecular structures of chosen compounds, which were used as input data for two predictive models including Traditional Artificial Neural Network and the Fuzzy Artificial Neural Network. They were trained to predict important physicochemical properties like boiling point, molar refractivity, hydrogen bond donors and acceptors, lipophilicity and others affecting drug effectiveness. The models performance was assessed through statistical measures such as Mean Squared Error, Root Mean Squared Error, Mean Absolute Error and the coefficient of determination (R \(^2\) ). The obtained results demonstrate that the integration of eccentricity based topological descriptors with the Fuzzy ANN model improved predictive accuracy compared with the Traditional ANN model, with \(R^{2}\) values reaching up to 0.977482, while for selected properties such as density, the \(R^{2}\) value improved from 0.890440 to 0.897012, confirming the effectiveness of fuzzy learning in modeling uncertainty and nonlinear molecular relationships. This research highlights the promise of topological indices combined with machine learning strategies in enabling the rational design and drug optimization for Traumatic Brain Injury.