Comparative Evaluation of Neural Network and Agent-Based Models for Organic Chemistry Reaction Predictions

The paper investigates the efficacy of two distinct computer simulation methods—Neural Network and Agent-Based Model (ABM)—for predicting basic organic chemistry reactions. We compare these methods based on their complexity, accuracy, and generalizability in predicting two fundamental reaction mechanisms: Elimination and Addition reactions. The study initially tests both models on an alkene halogen addition reaction and further examines their ability to generalize to elimination reactions. The Neural Network approach employs Long Short-Term Memory (LSTM) networks, utilizing SMILES representations of molecules to predict reaction outcomes. In contrast, the ABM simulates chemical interactions based on agent behaviors. Our results indicate that while the LSTM model accurately predicts basic reaction patterns and preserves atom counts, it encounters limitations with novel halogens and complex reaction variations. The ABM’s performance is evaluated relative to these findings, offering insights into the strengths and constraints of both modeling approaches in computational chemistry.