Scholarlink

<p>Molecular property prediction&#xa0;(MPP) plays a pivotal role in a wide range of domains, spanning drug discovery, material science, and environmental chemistry. With the rapid expansion of chemical data and advancements in artificial intelligence, recent years have seen significant progress in MPP. Due to the complex nature of molecular data such as molecular structures and SMILES notation, representation learning has emerged as a crucial strategy in extracting meaningful and interpretable features. This review examines recent single and multimodal techniques that uses&#xa0; these molecular representations, and outline datasets and tools available for feature generation. While SMILES-based and graph-based methods currently dominate, the potential of integrating multiple modalities remains underexplored. The key areas that require further investigation include model generalization and transferability across diverse chemical domains, improving interpretability, and integrating heterogeneous molecular data. By identifying critical limitations and proposing targeted research pathways, this review aims to advance robust, interpretable, and scalable MPP methods to promote accelerated progress in drug discovery.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Advancements in Molecular Property Prediction: A Survey of Single and Multimodal Approaches

  • 期刊论文

  • 发表日期：

    2025年7月26日

查看全文
  • Tanya Liyaqat,
  • Tanvir Ahmad,
  • Chandni Saxena

摘要

Molecular property prediction (MPP) plays a pivotal role in a wide range of domains, spanning drug discovery, material science, and environmental chemistry. With the rapid expansion of chemical data and advancements in artificial intelligence, recent years have seen significant progress in MPP. Due to the complex nature of molecular data such as molecular structures and SMILES notation, representation learning has emerged as a crucial strategy in extracting meaningful and interpretable features. This review examines recent single and multimodal techniques that uses  these molecular representations, and outline datasets and tools available for feature generation. While SMILES-based and graph-based methods currently dominate, the potential of integrating multiple modalities remains underexplored. The key areas that require further investigation include model generalization and transferability across diverse chemical domains, improving interpretability, and integrating heterogeneous molecular data. By identifying critical limitations and proposing targeted research pathways, this review aims to advance robust, interpretable, and scalable MPP methods to promote accelerated progress in drug discovery.