Abstract

Generative artificial intelligence (AI) is revolutionizing chemistry by designing novel molecules,predicting reactions, and accelerating discovery. Advanced models-ranging from SMILES-based VAEs and transformers to graph neural networks and diffusion frameworks-learn from massive databases (e.g. PubChem, ChEMBL) to navigate chemical space (~1033 molecules), generate valid structures with tailored properties, and enforce chemical constraints (valency, stereochemistry). These approaches outperform traditional methods in tasks like retrosynthesis planning and molecular optimization, with experimental validation such as Al-designed inorganic crystals synthesized in the lab. Importantly, this work emphasizes sustainable molecular and materials design. Generative pipelines are being adapted to minimize environmental impact via green-by-design objectives such as reducing process mass intensity (PMI), selecting biodegradable or non-toxic alternatives, and optimizing the atom economy. Al-guided materials like ZIF-8 frameworks and MOFs have been discovered via electrochemical synthesis with significantly lower energy use and waste, demonstrating eco-efficient design. Models are also used to design solvents and catalysts with improved environmental profiles. Such sustainability-aware Al tools support greener drug and materials development by integrating life-cycle thinking directly into molecular generative workflows.

References

• A. Lavecchia,“Navigating the frontier of drug‑like chemical space with cutting‑edge generative AI models,” Drug Discov. Today, vol. 29, no. 2, p. 103712, Feb. 2024. doi:10.1016/j.drudis.2023.103712
• P. Schwaller et al., “Predicting retrosynthetic pathways using transformer‑based models and a hyper‑graph exploration strategy, ” Chem. Sci., vol. 11, no. 12, pp. 3316–3325, 2020. doi:10.1039/C9SC05704H
• C. Zeni et al., “A generative model for inorganic materials design, ” Nature, vol. 626, no. 7997, pp. 600–607, Feb. 2025. doi:10.1038/s41586-025-08628-5
• J. Amabilino et al., “Guiding inverse molecular design with deep generative models: A review of architectures and applications, ” Chem. Sci., vol. 13, no. 29, pp. 8423–8439, 2022. doi:10.1039/D2SC01477A
• L. Wang et al., arXiv:2502.09511, Feb. 2025. “Diffusion models for molecules: A survey of methods and tasks, ” arXiv preprint
• A. N. Bradshaw et al., “Transformer‑based retrosynthetic planning: Improving performance through model scaling, ” J. Chem. Inf. Model., vol. 60, no. 1, pp. 11–20, Jan. 2020. doi:10.1021/acs.jcim.9b00845
• M. A. Masood, S. Kaski, and T. Cui, “Molecular property prediction using pretrained‑BERT and Bayesian active learning: A data‑efficient approach to drug design, ” J. Cheminform., vol. 17, no. 1, p. 58, 2025. doi:10.1186/s13321-025-00986-6
• Y . Wu, A. Walsh, and A. M. Ganose, “Race to the bottom: Bayesian optimization for chemical problems, ” Digit. Discov., vol. 3, no. 3, pp. 452–467, 2024. doi:10.1039/D3DD00234A
• D. Polykovskiy et al., “Molecular sets (MOSES): A benchmarking platform for molecular generation models, ” Front. Pharmacol., vol. 11, p. 565644, 2021. doi:10.3389/fphar.2020.565644
• G. Szymański et al., “Critical assessment of synthetic accessibility scores in computer‑assisted synthesis planning, ” J. Cheminform., vol. 15, no. 1, p. 35, 2023. doi:10.1186/s13321-023-00678-z
• W. Jin, R. Barzilay, and T. Jaakkola, “Junction tree variational autoencoder for molecular graph generation, ” in Proc. 35th Int. Conf. Mach. Learn. (ICML), Stockholm, Sweden, 2018, pp. 2323–2332.
• N. De Cao and T. Kipf, arXiv:1805.11973, May 2018. “MolGAN: An implicit generative model for small molecular graphs, ” arXiv preprint
• N. Madhawa, K. Yan, and Y . Bengio, arXiv preprint arXiv:1905.11600, May 2019. “GraphNVP: An invertible flow model for generating molecular graphs, ”
• J. Ross et al., “Large‑scale chemical language representations capture molecular structure and properties,” Nat. Mach. Intell., vol. 4, no. 6, pp. 519–532, Jun. 2022. doi:10.1038/s42256-022-00491-7
• S. Chithrananda, G. Grand, and B. Ramsundar, “ChemBERTa: Large‑scale self‑supervised pretraining for molecular property prediction, ” arXiv preprint arXiv:2010.09885, Oct. 2020.
• C. Edwards et al., “Translation between molecules and natural language, ” in Proc. 2022 Conf. Empir. Methods Nat. Lang. Process. (EMNLP), Abu Dhabi, UAE, Dec. 2022, pp. 375–413. doi:10.18653/v1/2022.emnlp-main.26
• M. H. S. Segler, T. Kogej, C. Tyrchan, and M. P. Waller, “Generating focused molecule libraries for drug discovery with recurrent neural networks, ” ACS Cent. Sci., vol. 3, no. 1, pp. 43–50, Jan. 2017. doi:10.1021/acscentsci.6b00362
• C. W. Coley et al., “Computer‑assisted retrosynthesis based on molecular similarity, ” ACS Cent. Sci., vol. 5, no. 9, pp. 1572–1583, Sep. 2019. doi:10.1021/acscentsci.9b00744
• Y . Guo et al., “RetroExplainer: Interpretable retrosynthesis with molecular assembly, ” Nat. Commun., vol. 14, no. 1, p. 6150, Oct. 2023. doi:10.1038/s41467-023-41698-5
• T. Blaschke, J. Arús-Pous, and O. Engkvist, “Geometry‑based scoring of de novo generated molecules for synthetic accessibility, ” J. Chem. Inf. Model., vol. 60, no. 1, pp. 75–89, Jan. 2020. doi:10.1021/acs.jcim.9b00839.
• J. S. Smith et al., “Accelerating material design with the generative toolkit for scientific discovery,“ npj Comput. Mater., vol. 9, no. 1, p. 69, May 2023. doi:10.1038/s41524-023-01028-1