Back

Multiomic foundation model predicts epigenetic regulation by zero-shot

Yang, Z.; Fan, X.; Lan, M.; Li, X.; You, Y.; Tian, L.; Church, G.; Liu, X.; Gu, F.

2024-12-22 genomics
10.1101/2024.12.19.629561 bioRxiv
Show abstract

Deciphering cell type-specific perturbation effects on genes and cellular states demands considerable experimental resources. To overcome this challenge, we propose Perturbation Transformer (PertFormer), a foundation model to uncover functional and regulatory mechanisms through interpretable in silico perturbation process. PertFormer comprises 3 billion parameters pretrained on both bulk and single-cell datasets covering 9 types of multiomics (55 billion bp bulk multiomics and the largest-to-date 1.5 billion paired multiomic samples from 1 million single cells), capturing regulation of 300 kb around every genic region. PertFormer enables zero-shot prediction of functional perturbations and cell fate dynamics, outperforming existing methods by 20.9%-480.2%. PertFormer identifies novel tumor treatment targets, validated experimentally. The generalization capabilities of PertFormer have the potential to accelerate the discovery of biological and clinical targets.

Matching journals

The top 3 journals account for 50% of the predicted probability mass.

1
Nature Communications
○ 5641 papers in training set
Top 2%
33.9%
Similar papers in this journal
2
Nature Methods
○ 385 papers in training set
Top 1%
9.6%
Similar papers in this journal
3
Genome Biology
○ 637 papers in training set
Top 1%
7.8%
Similar papers in this journal
50% of probability mass above
4
Cell Genomics
○ 172 papers in training set
Top 0.5%
6.7%
Similar papers in this journal
5
Nature Genetics
○ 286 papers in training set
Top 2%
4.0%
Similar papers in this journal
6
Genome Research
● 468 papers in training set
Top 2%
3.2%
7
Nature Machine Intelligence
○ 70 papers in training set
Top 0.8%
3.2%
8
Nucleic Acids Research
◐ 1281 papers in training set
Top 6%
3.2%
9
Nature Biotechnology
○ 172 papers in training set
Top 1%
3.2%
10
Nature
○ 645 papers in training set
Top 5%
2.4%
11
eLife
● 5828 papers in training set
Top 48%
1.9%
12
Molecular Systems Biology
○ 162 papers in training set
Top 1%
1.7%
13
Communications Biology
○ 993 papers in training set
Top 17%
1.5%
14
Molecular Cell
○ 350 papers in training set
Top 4%
1.1%
15
Cell Reports
○ 1498 papers in training set
Top 25%
1.0%
16
Nature Computational Science
○ 55 papers in training set
Top 1%
1.0%
17
Cell
○ 431 papers in training set
Top 10%
0.8%
18
Science Advances
● 1243 papers in training set
Top 31%
0.8%
19
Cell Systems
○ 201 papers in training set
Top 5%
0.8%
20
Proceedings of the National Academy of Sciences
● 2444 papers in training set
Top 42%
0.8%
21
NAR Genomics and Bioinformatics
◐ 242 papers in training set
Top 4%
0.8%
22
Cell Reports Methods
○ 165 papers in training set
Top 4%
0.6%
23
Briefings in Bioinformatics
◐ 354 papers in training set
Top 8%
0.6%
24
Nature Structural & Molecular Biology
236 papers in training set
Top 4%
0.5%

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.