Beyond performance: How design choices shape chemical language models

Fuente: Zenodo
Saved in:
Bibliographic Details
Main Authors: Fender, Inken, Gut, Jannik, Lemmin, Thomas
Format: Recurso digital
Published: Zenodo 2025
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866902186258071552
author Fender, Inken
Gut, Jannik
Lemmin, Thomas
author_facet Fender, Inken
Gut, Jannik
Lemmin, Thomas
contents <h2>File content</h2> <p><span>Pre-trained fairseq models and tokenizers for the 16 combinations of SMILES/SELFIES chemical language, atomwise/SentencePiece tokenizer, implicit/explict chirality representations, and  BART/RoBERTa model architecture.</span></p> <h2>Abstract</h2> <p>Chemical language models (CLMs) have shown strong performance in molecular property prediction and generation tasks. However, the impact of design choices, such as molecular representation format, tokenization strategy, and model architecture, on both performance and chemical interpretability remains underexplored. In this study, we systematically evaluate how these factors influence CLM performance and chemical understanding. We evaluated models through finetuning on downstream tasks and probing the structure of their latent spaces using simple classifiers and dimensionality reduction techniques. Despite similar performance on downstream tasks across model configurations, we observed substantial differences in the structure and interpretability of their internal representations. SMILES molecular representation format with atomwise tokenization strategy consistently produced more chemically meaningful embeddings, while models based on BART and RoBERTa architectures yielded comparably interpretable representations. These findings highlight that design choices meaningfully shape how chemical information is represented, even when external metrics appear unchanged. This insight can inform future model development, encouraging more chemically grounded and interpretable CLMs.</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_16926537
institution Zenodo
language
publishDate 2025
publisher Zenodo
record_format zenodo
spellingShingle Beyond performance: How design choices shape chemical language models
Fender, Inken
Gut, Jannik
Lemmin, Thomas
<h2>File content</h2> <p><span>Pre-trained fairseq models and tokenizers for the 16 combinations of SMILES/SELFIES chemical language, atomwise/SentencePiece tokenizer, implicit/explict chirality representations, and  BART/RoBERTa model architecture.</span></p> <h2>Abstract</h2> <p>Chemical language models (CLMs) have shown strong performance in molecular property prediction and generation tasks. However, the impact of design choices, such as molecular representation format, tokenization strategy, and model architecture, on both performance and chemical interpretability remains underexplored. In this study, we systematically evaluate how these factors influence CLM performance and chemical understanding. We evaluated models through finetuning on downstream tasks and probing the structure of their latent spaces using simple classifiers and dimensionality reduction techniques. Despite similar performance on downstream tasks across model configurations, we observed substantial differences in the structure and interpretability of their internal representations. SMILES molecular representation format with atomwise tokenization strategy consistently produced more chemically meaningful embeddings, while models based on BART and RoBERTa architectures yielded comparably interpretable representations. These findings highlight that design choices meaningfully shape how chemical information is represented, even when external metrics appear unchanged. This insight can inform future model development, encouraging more chemically grounded and interpretable CLMs.</p>
title Beyond performance: How design choices shape chemical language models
url https://doi.org/10.5281/zenodo.16926537