SELFIES-TED : A Robust Transformer Model for Molecular Representation using SELFIES
Keywords: molecular representation, property prediction, molecular generation
Abstract: Large-scale molecular representation methods have revolutionized applications in material science, such as drug discovery, chemical modeling, and material design. With the rise of transformers, models now learn representations directly from molecular structures. In this paper, we introduce SELFIES-TED, a transformer-based model designed for molecular representation using SELFIES, a more robust, unambiguous method for encoding molecules compared to traditional SMILES strings. By leveraging the robustness of SELFIES and the power of the transformer encoder-decoder architecture, SELFIES-TED effectively captures the intricate relationships between molecular structures and their properties. Having pretrained with 1 billion molecule samples, our model demonstrates improved performance on molecular property prediction tasks across various benchmarks, showcasing its generalizability and robustness.
Additionally, we explore the latent space of SELFIES-TED, revealing valuable insights that enhance its capabilities in both molecule property prediction and molecule generation tasks, opening new avenues for innovation in molecular design.
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 8268
Loading