Related Experiment Video
Updated: Jun 26, 2025

Modeling an Enzyme Active Site using Molecular Visualization Freeware
Published on: December 25, 2021
Rethinking the Masking Strategy for Pretraining Molecular Graphs from a Data-Centric View.
Wei Lin1, Chi Chung Alan Fung1
1Department of Neuroscience, City University of Hong Kong, Tat Chee Avenue, Kowloon Tong, Kowloon 999077, Hong Kong, China.
A new weighted masking strategy improves molecular graph pretraining by addressing imbalanced atom distribution. This method enhances model capacity and performance without adding parameters, outperforming random masking.
Area of Science:
- Computational chemistry
- Machine learning
- Drug discovery
Background:
- Node-level self-supervised learning is crucial for pretraining molecular graphs.
- Attribute Masking (AttrMask) is a key method, with enhancements focusing on model architecture.
- Existing methods neglect imbalanced atom distribution from random masking.
Purpose of the Study:
- To introduce a weighted masking strategy for more effective molecular graph pretraining.
- To enhance the capacity of pretrained models by better utilizing molecular information.
- To improve performance over random masking without increasing model parameters.
Main Methods:
- Proposed a novel weighted masking strategy tailored to molecular properties.
- Applied the strategy to the Attribute Masking (AttrMask) framework.
- Evaluated performance against random masking and model-centric approaches.
Main Results:
- The weighted masking strategy significantly improved performance compared to random masking.
- Performance gains were achieved without increasing model parameters.
- The strategy demonstrated effectiveness and potential for broader application.
Conclusions:
- Weighted masking is a superior alternative to random masking for molecular graph pretraining.
- This approach enhances model capacity and performance by addressing atom distribution.
- The proposed strategy is adaptable to other pretraining methods for improved results.
Related Concept Videos
Masking and Demasking Agents
There are many masking agents, such as cyanide, fluoride, triethanolamine, thiourea, and 2,3-bis(sulfanyl)propan-1-ol (formerly 2,3-dimercapto-1-propanol), with the masking agent chosen based on...
Fischer Projections
Molecular Models
Newman Projections
The organic molecules rotate across the single bonds leading to numerous temporary three-dimensional structures of varying energy known as...
¹³C NMR: Distortionless Enhancement by Polarization Transfer (DEPT)
Protecting Groups for Aldehydes and Ketones: Introduction

