MetaMolGen: A Neural Graph Motif Generation Model for De Novo Molecular Design

2025-04-22

AI Summary

To address data scarcity and poor conditional generalization in molecular generation for drug discovery, this work proposes a few-shot, property-controllable de novo molecular design framework. Methodologically, it introduces first-order model-agnostic meta-learning (MAML) to molecular generation for the first time; proposes a graph-motif-normalized latent-space mapping mechanism to enhance structural fidelity; and designs a learnable, differentiable property projector for end-to-end conditional control. Evaluated under few-shot settings—requiring only a handful of target property examples—the framework achieves >98% validity, significantly outperforms baselines in diversity, and rapidly adapts to unseen property tasks. Moreover, it enables efficient inverse molecular design. By unifying meta-learning with structured latent-space regularization and differentiable property conditioning, the framework establishes a novel paradigm for controllable molecular generation in low-data regimes.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Search and Optimization: Metareasoning and Metaheuristics
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: ML for personalized search and recommendations

Abstract

Molecular generation plays an important role in drug discovery and materials science, especially in data-scarce scenarios where traditional generative models often struggle to achieve satisfactory conditional generalization. To address this challenge, we propose MetaMolGen, a first-order meta-learning-based molecular generator designed for few-shot and property-conditioned molecular generation. MetaMolGen standardizes the distribution of graph motifs by mapping them to a normalized latent space, and employs a lightweight autoregressive sequence model to generate SMILES sequences that faithfully reflect the underlying molecular structure. In addition, it supports conditional generation of molecules with target properties through a learnable property projector integrated into the generative process.Experimental results demonstrate that MetaMolGen consistently generates valid and diverse SMILES sequences under low-data regimes, outperforming conventional baselines. This highlights its advantage in fast adaptation and efficient conditional generation for practical molecular design.

Problem

• Addresses few-shot molecular generation in drug discovery
• Standardizes graph motifs via normalized latent space mapping
• Enables conditional generation with target property control

Innovation

• First-order meta-learning for few-shot generation
• Normalized latent space for graph motifs
• Learnable property projector for conditional generation