GraphBioisostere: General Bioisostere Prediction Model with Deep Graph Neural Network
Sho MASUNAGA *, Kairi FURUI, Apakorn KENGKANNA, Masahito OHUE
Department of Computer Science, School of Computing, Institute of Science Tokyo
[Purpose]
Lead optimization to improve pharmacokinetics and toxicity while maintaining biological activity is an important and costly stage in the drug discovery process, requiring computational approaches for increased efficiency. Bioisosteric replacement is known to be an effective method for lead optimization, and its prediction is considered important for streamlining the drug discovery process.
[Methods]
We propose GraphBioisostere, a bioisostere prediction model that uses graph neural networks. The proposed model leverages a large-scale matched molecular pair dataset constructed from the ChEMBL database and directly learns bioisosterism without target information by considering entire chemical structures.
[Results and Discussion]
Our evaluation demonstrates that GraphBioisostere outperforms conventional LightGBM-based methods, and models directly learning potency changes failed to discriminate bioisosteres, highlighting the importance of specifically learning bioisosterism.
Additionally, models pre-trained on target-independent bioisostere prediction exhibited excellent transfer learning performance for potency change prediction against specific targets. This suggests that GraphBioisostere acquires useful representations of the relationship between chemical structure and activity.
[Conclusions]
Our research provides a tool to evaluate the potential of structural changes in molecular pairs to maintain activity independently of targets, contributing to improved efficiency in the drug discovery process.