Developing and evaluating Bag of Smiles and ChemGrams for chemical data tokenization

Abstract

This thesis introduces two novel tokenisation algorithms, Bag of Smiles and ChemGrams, which are designed to revolutionize the field of molecular representation and predictive performance in the context of corrosion inhibitor datasets. These algorithms, developed based on data extracted from academic papers focusing on the efficiency of corrosion inhibitors, offer a fresh perspective by capturing critical chemical structures and contextual interactions within molecules that traditional SMILES-based tokenization methods often overlook.