[Paper Review] Molecular Representations for AI in Chemistry and Materials Science: An NLP Perspective

Sanjanasri JP, Pratiti Bhadra+2 more|arXiv (Cornell University)|Feb 24, 2026

Machine Learning in Materials Science0 citations

TL;DR

The paper surveys NLP-inspired molecular representations (string-based and graph-based) for AI in chemistry and materials science, highlighting strengths, weaknesses, and downstream AI applications.

ABSTRACT

Deep learning, a subfield of machine learning, has gained importance in various application areas in recent years. Its growing popularity has led it to enter the natural sciences as well. This has created the need for molecular representations that are both machine-readable and understandable to scientists from different fields. Over the years, many chemical molecular representations have been constructed, and new ones continue to be developed as computer technology advances and knowledge of molecular complexity increases. This paper presents some of the most popular digital molecular representations inspired by natural language processing (NLP) and used in chemical informatics. In addition, the paper discusses some notable AI-based applications that use these representations. This paper aims to provide a guide to structural representations that are important for the application of AI in chemistry and materials science from the perspective of an NLP researcher. This review is a reference tool for researchers with little experience working with chemical representations who wish to work on projects at the interface of these fields.

Motivation & Objective

• Motivate the need for machine-readable molecular representations for AI-driven chemistry and materials research.
• Survey widely used string- and graph-based representations and their NLP-inspired origins.
• Explain challenges in representing molecules for machine learning and how different representations address them.
• Provide guidance for NLP researchers entering cheminformatics and materials science AI tasks.

Proposed method

• Review and categorize molecular representations into string-based and graph-based approaches.
• Discuss SMILES, InChI, DeepSMILES, SELFIES, and their respective limitations and enhancements.
• Explain graph-based matrix representations and their use in AI tasks.
• Describe downstream applications and models inspired by NLP concepts (e.g., Mol2vec, Smiles2vec, Graph2SMILES).

Figure 1: Two Dimensional and Three Dimensional Structure of 3,4-Methylenedioxymethamphetamine (MDMA) molecule

Experimental results

Research questions

• RQ1 What NLP-inspired representations are most effective for modeling chemical and material structures with AI?
• RQ2 How do different representations handle 3D structure, stereochemistry, and chemical validity in ML tasks?
• RQ3 What are the trade-offs between string-based and graph-based representations for downstream AI applications in chemistry and materials science?

Key findings

• SMILES enables NLP-like embeddings but suffers from ambiguity, limited expressiveness for 3D structure, and semantic/syntactic errors.
• InChI provides a standardized, layered, machine-readable identifier but can be long and lacks explicit hydrogen representation; InChI Keys offer a compact alternative.
• DeepSMILES and SELFIES address some SMILES limitations, with SELFIES guaranteeing chemical validity and reducing syntactic/semantic errors.
• Graph-based representations (adjacency, distance, connectivity matrices) offer flexible, information-rich structures better suited for many AI tasks; graph encoders and transfer learning enable cross-task reuse.
• Mol2vec and related embedding techniques treat molecular fragments as words to produce informative embeddings for property prediction and screening.
• Graph2SMILES and transformer-based graph encoders demonstrate advances in bridging graph representations with SMILES-like outputs for generative tasks.

Figure 2: Connectivity Matrix Representation for the MDMA Molecule