A deep-learning system bridging molecule structure and biomedical text with comprehension comparable to human professionals

Abstract

To accelerate biomedical research process, deep-learning systems are developed to automatically acquire knowledge about molecule entities by reading large-scale biomedical data. Inspired by humans that learn deep molecule knowledge from versatile reading on both molecule structure and biomedical text information, we propose a knowledgeable machine reading system that bridges both types of information in a unified deep-learning framework for comprehensive biomedical research assistance. We solve the problem that existing machine reading models can only process different types of data separately, and thus achieve a comprehensive and thorough understanding of molecule entities. By grasping meta-knowledge in an unsupervised fashion within and across different information sources, our system can facilitate various real-world biomedical applications, including molecular property prediction, biomedical relation extraction and so on. Experimental results show that our system even surpasses human professionals in the capability of molecular property comprehension, and also reveal its promising potential in facilitating automatic drug discovery and documentation in the future.

Unraveling the mystery of efficacy in Chinese medicine formula: New approaches and technologies for research on pharmacodynamic substances

“…For ADME prediction, it has been used in studies of rutin ( Liu et al, 2021a ), cardioplegia capsules ( Guo et al, 2018 ), and Chrysoeriol ( Liu et al, 2021b ) to screen bioactive compounds and reveal PK properties. In terms of deep machine learning, it has been used to design drug targets and new drug discovery now commonly combine machine learning and deep learning algorithms to improve the efficiency and quality of the development output ( Zeng et al, 2022 , Li et al, 2022b , He et al, 2019 ). Similar strategies applied in classical Chinese herbal formulas ( Yang et al, 2020 ), pediatric poria granules ( Guo et al, 2020 ) and gentian violet ( Wang et al, 2021a ).…”

A Systematic Survey of Molecular Pre-trained Models

“…Inspired by the advances of multimodal pre-training in computer vision and NLP domains [74,77], some recent works perform multimodal pre-training on molecules. For example, KV-PLM [129] first tokenizes both the SMILES strings and biochemical texts. Then, they randomly mask part of the tokens and pre-train the neural encoders to recover the masked tokens.…”

“…The abovementioned multimodal pre-training strategies can advance the performance in the translations between various modalities. Additionally, various modalities can complement with each other and constitute a more complete knowledge base for downstream tasks [129].…”

“…Currently, MPMs have already shown promising values in molecular generation. For example, MolGPT [4] adopts the autoregressive [21,129] enables generation molecules from a descriptive text (textgeneration). In addition, molecular conformation generation which aims to generate 3D conformations of molecules has wide applications such as protein-ligand binding pose prediction [24,121].…”

A Machine Learning Method for Material Property Prediction: Example Polymer Compatibility

“…Nevertheless, numerical algorithms and computing resources become the main bottleneck when it comes to macroscale and multibody system. Machine Learning (ML) which is data-driven achieves great progress on various tasks, such as image recognition [4,5] and natural language processing [6,7]. Recently, scientists start to apply machine learning to material research and property prediction.…”