Multi-stage variational autoencoders for hierarchical molecular generation and activity optimization
- 12 November 2025
- journal article
- research article
- Published by Springer Nature in Journal of Computer-Aided Molecular Design
- Vol. 39 (2) , 1-20
Abstract
Deep generative models may detect novel compounds with favourable features, exhibiting chemical design potential. Traditional single-stage variational autoencoders (VAEs) lack validity, uniqueness, and biologically meaningful distribution alignment. It is difficult to represent global molecular architecture and chemical properties in a single latent representation. To overcome these challenges, we offer a multi-stage VAE system that encodes and decodes molecular representations in sequence. Improvements to latent space retain structural integrity while also adding innovation and distinction. Validity, originality, novelty, Fréchet ChemNet Distance (FCD), and KL divergence are used to validate the methodology with ChEMBL and polymer datasets. The bioefficacy of EGFR inhibitors is evaluated using computational Chemprop-based QSAR models. We offer adaptive fine-tuning strategies for the inner-layer (IL) and outer-layer (OL) to improve generating accuracy. IL adaptability is most suited to active compounds. Quantitative evaluations indicate consistent gains in validity, novelty, and biological activity over strong baselines (for example, MoLeR and RationaleRL). We give MNIST tests that confirm the hierarchical training method’s stability but not its scalability beyond molecular tasks, ensuring cross-domain applicability. For generative drug discovery, hierarchical latent models with a multi-stage VAE are advised.Keywords
This publication has 28 references indexed in Scilit:
- Transform Drug Discovery and Development With Generative Artificial Intelligence2025
- An Effective Pipeline for Training Variational Autoencoders for Synthesizable and Optimized Molecular DesignIEEE Access, 2024
- Generative AI for Transformative Healthcare: A Comprehensive Study of Emerging Models, Applications, Case Studies, and LimitationsIEEE Access, 2024
- MedMetaverse: Medical Care of Chronic Disease Patients and Managing Data Using Artificial Intelligence, Blockchain, and Wearable Devices State-of-the-Art MethodologyIEEE Access, 2023
- Generative and reinforcement learning approaches for the automated de novo design of bioactive compoundsCommunications Chemistry, 2022
- Inverse design of 3d molecular structures with conditional generative neural networksNature Communications, 2022
- Generative Adversarial Neural Network for Creating Photorealistic Images2020
- ChEMBL: towards direct deposition of bioassay dataNucleic Acids Research, 2018
- Application of Generative Autoencoder in De Novo Molecular DesignMolecular Informatics, 2017
- Testing the manifold hypothesisJournal of the American Mathematical Society, 2016