DTA

Corso Ordinario Secondo Livello

RONCOLI, ANDREA

etd-10102025-173022

Transfer Learning of Molecular Representations: Strings to Graphs

Classe Scienze Sperimentali

INGEGNERIA - INGEGNERIA

Tutor Prof. STEFANINI, CESARE
Relatore Prof. AVIZZANO, CARLO ALBERTO

• chemical language models
• molecular machine learning
• transfer learning

12/12/2025;

completa

This work explores the potential of transfer learning between different molecular representations by leveraging pretrained language models on SMILES strings to enhance downstream graph-based machine learning models. We investigate whether representations learned in the sequence (string) domain can provide useful inductive biases for models operating on molecular graphs, thereby bridging two common paradigms in molecular representation learning. To enable this cross-representation transfer, we develop a method to align SMILES tokens with corresponding graph atoms, allowing pretrained embeddings from language models to be integrated as node features in graph neural networks.

Nome file	Dimensione
Tesina_R...i_Def.pdf	1.35 Mb
Contatta l'autore