A Lightweight, Integrated Generative AI Assistant for Accelerated Early-Stage Drug Discovery on Constrained-Resource Hardware

This repository contains the official replication data, code, and trained models for the manuscript: "A Lightweight, Integrated Generative AI Assistant for Accelerated Early-Stage Drug Discovery on Constrained-Resource Hardware." Contents Include: Datasets: Custom train/validation/test splits, PAINS structural safety filters, and 1,000 newly generated novel molecules (SMILES) evaluated in the study. Models & Code: Trained model weights (.pt, .pkl) and Python evaluation scripts for the LSTM generator, XGBoost toxicity/solubility predictors, and the Streamlit application (optimized for 4GB VRAM GPUs). Note: The original raw Delaney (ESOL) and Tox21 datasets utilized in this study are publicly available. Full provenance, DOIs, and download links for these original datasets are detailed in the included README.md file.