Back

3D Molecular Pretraining via Localized Geometric Generation

Sun, Y.; Chen, K.; Liu, K.; Ye, Q.

2024-09-14 bioinformatics
10.1101/2024.09.10.612249 bioRxiv
Show abstract

Self-supervised learning on 3D molecular structures is gaining importance in data-driven scientific research and applications due to the high costs of annotating bio-chemical data. However, the strategic selection of semantic units for modeling 3D molecular structures remains underexplored, despite its crucial role in effective pre-training--a concept well-established in language processing and computer vision. We introduce Localized Geometric Generation (LEGO), a novel approach that treats tetrahedrons within 3D molecular structures as fundamental building blocks, leveraging their geometric simplicity and widespread presence across chemical functional patterns. Inspired by masked modeling, LEGO perturbs tetrahedral local structures and learns to reconstruct them in a self-supervised manner. Experimental results demonstrate LEGO consistently enhances molecular representations across biochemistry and quantum property prediction benchmarks. Additionally, the tetrahedral modeling and pretraining generalize from small molecules to larger molecular systems, validating by protein-ligand affinity prediction. Our results highlight the potential of selecting semantic units to build more expressive and interpretable neural networks for scientific AI applications.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

1
Nature Machine Intelligence
61 papers in training set
Top 0.1%
26.7%
2
Advanced Science
249 papers in training set
Top 1%
10.4%
3
Nature Communications
4913 papers in training set
Top 27%
6.6%
4
Bioinformatics
1061 papers in training set
Top 4%
5.0%
5
Proceedings of the National Academy of Sciences
2130 papers in training set
Top 19%
3.8%
50% of probability mass above
6
Scientific Reports
3102 papers in training set
Top 43%
2.8%
7
Computational and Structural Biotechnology Journal
216 papers in training set
Top 3%
2.4%
8
Communications Biology
886 papers in training set
Top 5%
2.1%
9
Journal of Chemical Information and Modeling
207 papers in training set
Top 2%
1.7%
10
Patterns
70 papers in training set
Top 0.8%
1.7%
11
Nucleic Acids Research
1128 papers in training set
Top 10%
1.7%
12
iScience
1063 papers in training set
Top 17%
1.5%
13
IEEE Transactions on Computational Biology and Bioinformatics
17 papers in training set
Top 0.3%
1.4%
14
Bioinformatics Advances
184 papers in training set
Top 3%
1.4%
15
PLOS Computational Biology
1633 papers in training set
Top 18%
1.4%
16
Nature Methods
336 papers in training set
Top 5%
1.3%
17
PLOS ONE
4510 papers in training set
Top 61%
1.1%
18
Frontiers in Immunology
586 papers in training set
Top 6%
1.1%
19
Briefings in Bioinformatics
326 papers in training set
Top 5%
1.0%
20
Medical Image Analysis
33 papers in training set
Top 0.8%
0.9%
21
Frontiers in Genetics
197 papers in training set
Top 8%
0.9%
22
npj Digital Medicine
97 papers in training set
Top 3%
0.8%
23
Cell Systems
167 papers in training set
Top 11%
0.8%
24
Nature
575 papers in training set
Top 14%
0.8%
25
Nature Biomedical Engineering
42 papers in training set
Top 2%
0.8%
26
Physical Review Research
46 papers in training set
Top 0.8%
0.8%
27
Science Bulletin
22 papers in training set
Top 0.8%
0.8%
28
Cell Research
49 papers in training set
Top 2%
0.8%
29
Journal of Molecular Biology
217 papers in training set
Top 3%
0.8%
30
Cell Genomics
162 papers in training set
Top 6%
0.8%