Predicting molecular properties is a key challenge in cheminformatics, essential for various applications in chemistry and drug discovery, and its success depends on generating informative molecular graphs. One key property in drug molecule discovery is lipophilicity, often measured as log P, which determines how easily a molecule dissolves in fat compared to water and plays a crucial role in drug absorption, distribution, metabolism, and excretion [1-2].

While quantum mechanical simulations provide highly accurate predictions of molecular properties, they are often computationally expensive. To overcome this limitation, researchers can leverage existing molecular drug databases [3] and utilize Graph Neural Networks (GNNs) [4], which offer an efficient way to learn molecular representations directly from structural data. Bande et al. have demonstrated the potential of GNNs for various molecular property predictions in computational chemistry [5-6].

Recent advancements in GNNs have enabled more effective modeling of molecular structures and accurate prediction of lipophilicity can accelerate drug development by reducing experimental costs [7-8]. In this study, we compare the performance of three GNN-based models—Graph Convolutional Network (GCN), Graph Isomorphism Network (GIN), and Attentive Fingerprint (AFP)—to predict lipophilicity for drug molecules [9-11]. We evaluate these models on publicly available datasets and compare their performance using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-squared (R2) metrics. Our results show that AFP outperforms GCN and GIN, providing more accurate lipophilicity predictions. These findings highlight the strengths and limitations of each model, providing insights into the effectiveness of GNNs in molecular lipophilicity property prediction.

Keywords: Graph Neural Network, Lipophilicity, Drug Molecules, GCN, AFP, GIN

Reference:

1. Leo, A., et al. Chem. Rev. 71 (1971), 525–616.

2. Isert, C., et al. ACS Omega 8 (2023), 2046–2056.

3. Hersey, A., EMBL-EBI, Tech. Rep. (2015), Feb.

4. Scarselli, F., et al. IEEE Trans. Neural Netw. 20 (2009), 61–80.

5. Kotobi, A., et al. J. Am. Chem. Soc. 145 (2023), 22584–22598.

6. Singh, K., et al. J. Chem. Theory Comput. 18 (2022), 4408–4417.

7. Ahmad, W., et al. ACS Omega 8 (2023), 3236–3244.

8. De Carlo, A., et al. Pharmaceutics 16 (2024), 776.

9. Kipf, T. N., et al. arXiv preprint (2017), arXiv:1609.02907.

10. Xu, K., et al. arXiv preprint (2019), arXiv:1810.00826.

11. Xiong, Z., et al. J. Med. Chem. 63 (2019), 8749–8760.