Back

Single-cell foundation models identify shared and divergent transcriptomic signatures of aging across invertebrates and mammals

Al Amin, M. A. U.; Le, K.; Qin, H.

2026-07-27 genomics
10.64898/2026.07.24.740647 bioRxiv
Show abstract

Cross-species aging clocks have so far stayed within mammals. We asked whether a single model can predict age from single cells of fly, worm, mouse, and human, species separated by roughly 800 million years of evolution. We fine-tuned scGPT and Geneformer on 1.3 million such cells, using one parameter set per architecture. Both models classified cells as young, middle, or old, reaching 78.6% and 80.5% accuracy. SHAP attribution then showed where the two architectures agreed and where they diverged. scGPT ranked ribosomal proteins highest in every species, with RPL12 first throughout. Geneformer did the same in mouse and human but favored signaling, chromatin, and ubiquitin-ligase genes in fly and worm. Hiding all 51 ribosomal-protein genes at inference cost 22-28 percentage points of accuracy, against 0.3 points for size-matched random gene sets. Bootstrap resampling, five-fold cross-validation, and comparison with six published aging studies left the rankings largely unchanged. Age-associated signal is therefore accessible to a pooled cross-species model, but which genes carry it depends on how the model encodes expression.

Matching journals

The top 8 journals account for 50% of the predicted probability mass.

1
Nature Aging
○ 60 papers in training set
Top 0.2%
9.8%
Similar papers in this journal
2
Aging Cell
○ 165 papers in training set
Top 0.4%
7.9%
Similar papers in this journal
3
Nature Communications
○ 5641 papers in training set
Top 21%
7.9%
Similar papers in this journal
4
PLOS Computational Biology
● 1863 papers in training set
Top 5%
7.3%
Similar papers in this journal
5
eLife
● 5828 papers in training set
Top 16%
7.3%
Similar papers in this journal
6
GeroScience
○ 109 papers in training set
Top 0.5%
4.8%
7
Genome Biology
○ 637 papers in training set
Top 3%
3.4%
8
Genome Research
● 468 papers in training set
Top 2%
3.2%
50% of probability mass above
9
Proceedings of the National Academy of Sciences
● 2444 papers in training set
Top 16%
3.2%
10
Science
● 477 papers in training set
Top 3%
3.1%
11
Cell Reports
○ 1498 papers in training set
Top 15%
2.4%
12
Aging
○ 75 papers in training set
Top 0.7%
2.1%
13
Molecular Systems Biology
○ 162 papers in training set
Top 1%
2.1%
14
Nature Methods
○ 385 papers in training set
Top 4%
2.1%
15
Nature Medicine
○ 125 papers in training set
Top 1%
2.1%
16
Scientific Reports
○ 3612 papers in training set
Top 52%
1.7%
17
Cell Systems
○ 201 papers in training set
Top 3%
1.7%
18
Briefings in Bioinformatics
◐ 354 papers in training set
Top 5%
1.5%
19
Nature
○ 645 papers in training set
Top 8%
1.3%
20
Genome Biology and Evolution
◐ 338 papers in training set
Top 2%
1.3%
21
Nucleic Acids Research
◐ 1281 papers in training set
Top 11%
1.1%
22
Computational and Structural Biotechnology Journal
● 242 papers in training set
Top 5%
1.1%
23
Cell Genomics
○ 172 papers in training set
Top 3%
1.1%
24
PLOS ONE
● 5266 papers in training set
Top 53%
1.1%
25
PLOS Genetics
● 862 papers in training set
Top 10%
1.0%
26
NAR Genomics and Bioinformatics
◐ 242 papers in training set
Top 4%
1.0%
27
Genome Medicine
○ 183 papers in training set
Top 4%
1.0%
28
Frontiers in Genetics
○ 230 papers in training set
Top 5%
0.9%
29
BMC Biology
○ 265 papers in training set
Top 5%
0.8%
30
Communications Biology
○ 993 papers in training set
Top 30%
0.8%

"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.