Yasuhiro Yoshikai

Yasuhiro Yoshikai, Tadahaya Mizuno, Shumpei Nemoto et al.

Recent years have seen rapid development of descriptor generation based on representation learning of extremely diverse molecules, especially those that apply natural language processing (NLP) models to SMILES, a literal representation of molecular structure. However, little research has been done on how these models understand chemical structure. To address this black box, we investigated the relationship between the learning progress of SMILES and chemical structure using a representative NLP model, the Transformer. We show that while the Transformer learns partial structures of molecules quickly, it requires extended training to understand overall structures. Consistently, the accuracy of molecular property predictions using descriptors generated from models at different learning steps was similar from the beginning to the end of training. Furthermore, we found that the Transformer requires particularly long training to learn chirality and sometimes stagnates with low performance due to misunderstanding of enantiomers. These findings are expected to deepen the understanding of NLP models in chemistry.

Zehao Li, Yasuhiro Yoshikai, Shumpei Nemoto et al.

Understanding how chemical language models (CLMs) learn chemical meaning from molecular string representations, rather than only surface-level string patterns, is an important question in chemical representation learning and machine learning for chemistry. Chirality provides a demanding test case: enantiomers can differ greatly in pharmacological activity and toxicity, yet CLMs often struggle to distinguish chiral configurations reliably. Here we present Pan-CORE (Pan-Chemical Omniscale Representation Engine), a family of autoregressive Transformer-based encoder-decoder models for SMILES translation, and use high-temporal-resolution checkpoint analysis to investigate how chiral information is learned during training. Across all tested Pan-CORE variants, we observe a reproducible jump-up in which chiral-token accuracy rises abruptly after a long plateau, suggesting that chiral learning stagnation is not explained by model capacity alone and instead reflects the complexity of chiral constraints. Analyses of attention dynamics, residual-stream trajectories, and latent-space geometry support an encoder-centered mechanism in which chiral-token representations undergo transient destabilization and reconstruction, seen as a V-shaped drop and recovery in vector norm and directional stability, together with a clear reorganization of chiral molecular representations in the latent space. Encoder-decoder cross-evaluation further supports the encoder-centered nature of the transition, and targeted attention-head ablation identifies a small set of chiral-sensitive heads whose removal selectively reduces chiral-token accuracy even in the fully trained model. These findings show that SMILES translation can serve as a useful experimental system for mechanistic analysis of semantic emergence in CLMs, with implications for interpretable chemical representation learning.

Yasuhiro Yoshikai, Tadahaya Mizuno, Shumpei Nemoto et al.

Recently, molecule generation using deep learning has been actively investigated in drug discovery. In this field, Transformer and VAE are widely used as powerful models, but they are rarely used in combination due to structural and performance mismatch of them. This study proposes a model that combines these two models through structural and parameter optimization in handling diverse molecules. The proposed model shows comparable performance to existing models in generating molecules, and showed by far superior performance in generating molecules with unseen structures. Another advantage of this VAE model is that it generates molecules from latent representation, and therefore properties of molecules can be easily predicted or conditioned with it, and indeed, we show that the latent representation of the model successfully predicts molecular properties. Ablation study suggested the advantage of VAE over other generative models like language model in generating novel molecules. It also indicated that the latent representation can be shortened to ~32 dimensional variables without loss of reconstruction, suggesting the possibility of a much smaller molecular descriptor or model than existing ones. This study is expected to provide a virtual chemical library containing a wide variety of compounds for virtual screening and to enable efficient screening.