t-SMILES: A Scalable Fragment-based Molecular Representation Framework for De Novo Molecule Generation

Effective representation of molecules is a crucial factor affecting theperformance of artificial intelligence models. This study introduces aflexible, fragment-based, multiscale molecular representation framework calledt-SMILES (tree-based SMILES) with three code algorithms: TSSA (t-SMILES withShared Atom), TSDY (t-SMILES with Dummy Atom) and TSID (t-SMILES with ID). Itdescribes molecules using SMILES-type strings obtained by performing abreadth-first search on a full binary tree formed from a fragmented moleculargraph. Systematic evaluations using JTVAE, BRICS, MMPA, and Scaffold show thefeasibility to construct a multi-code molecular description system, wherevarious descriptions complement each other, enhancing the overall performance.Additionally, it exhibits impressive performance on low-resource datasets,whether the model is original, data augmented, or pre-training fine-tuned. Itsignificantly outperforms classical SMILES, DeepSMILES, SELFIES and baselinemodels in goal-directed tasks. Furthermore, it surpasses start-of-the-artfragment, graph and SMILES based approaches on ChEMBL, Zinc, and QM9.