Recent advances in the self-referencing embedded strings (SELFIES) library

Article Chemistry, Multidisciplinary

Group SELFIES: a robust fragment-based molecular string representation

Austin H. Cheng et al.

Summary: This article introduces Group SELFIES, a molecular string representation that utilizes group tokens to represent functional groups or entire substructures while ensuring chemical robustness. The advantages of capturing chemical motifs and flexibility are demonstrated in experiments, showing improvement in distribution learning of common molecular datasets. The article also highlights the improvement in the quality of generated molecules compared to regular SELFIES strings through random sampling of Group SELFIES strings.

DIGITAL DISCOVERY (2023)

Add to Collection

Article Chemistry, Multidisciplinary

Model agnostic generation of counterfactual explanations for molecules

Geemi P. Wellawatte et al.

Summary: A major challenge in deep learning in chemistry is the lack of interpretability, which hinders the deployment of AI models. Counterfactuals provide a rationale for model predictions and insights into chemical structures. This work demonstrates a universal model-agnostic approach that can explain predictions of any black-box model.

CHEMICAL SCIENCE (2022)

Add to Collection

Article Computer Science, Artificial Intelligence

SELFIES and the future of molecular string representations

Mario Krenn et al.

Summary: Artificial intelligence and machine learning have gained popularity in the field of chemistry and materials science, requiring a fluent chemical language. The traditional molecular string representation, SMILES, has limitations, but the introduction of SELFIES in 2020 has solved these issues and enabled new applications in chemistry. Looking ahead, 16 future projects for robust molecular representations are proposed.

PATTERNS (2022)

Add to Collection

Article Chemistry, Multidisciplinary

STOUT: SMILES to IUPAC names using neural machine translation

Kohulan Rajan et al.

Summary: STOUT is a tool for compound naming using deep-learning neural machine translation, capable of generating IUPAC names from SMILES strings and performing reverse translation. The system achieves high accuracy, with even incorrect predictions showing high similarity to the true compounds.

JOURNAL OF CHEMINFORMATICS (2021)

Add to Collection

Article Chemistry, Multidisciplinary

Beyond generative models: superfast traversal, optimization, novelty, exploration and discovery (STONED) algorithm for molecules using SELFIES

AkshatKumar Nigam et al.

Summary: The study introduces a new algorithm called STONED, which achieves performance comparable to deep generative models in the chemical space through interpolation and exploration without the need for large amounts of data and training time.

CHEMICAL SCIENCE (2021)

Add to Collection

Article Chemistry, Multidisciplinary