Back

Identifiable inverse optimal transport for single-cell state transitions

Dong, H.; Gong, Y.; Liu, Y.

2026-10-06 bioinformatics
10.64898/2026.10.01.756056 bioRxiv
Show abstract

Signed effects of target-state programmes are unidentifiable in balanced optimal transport because fixed endpoint distributions make their contributions constant across feasible plans. We introduce UOT-IOT, a semi-relaxed inverse transport framework that restores measurable targetmass responses and provides a rank condition for local identifiability. It recovers coefficients inground-truth benchmarks (correlation 0.997 versus 0.081 for balanced transport), transfers frozen directions to six unseen samples with specificity controls, and achieves the lowest transition error against seven methods under matched evaluation conditions. GH-IOT separates shared directions from context and lineage-history offsets, revealing opposing history-associated effects. IFN/HLA is associated with patient treatment response, whereas EMT relates to drug sensitivity and perturbation-induced state changes. PERSIST-IOT improves future population-composition prediction in an independent experiment. Together, these modules distinguish current state,transition propensity and future population outcomes while supporting testable biological interpretations.

Matching journals

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

1
Nature Communications
○ 5641 papers in training set
Top 4%
26.5%
Similar papers in this journal
2
Cell Systems
○ 201 papers in training set
Top 0.2%
11.0%
Similar papers in this journal
3
Cell Reports Methods
○ 165 papers in training set
Top 0.1%
6.7%
Similar papers in this journal
4
Genome Biology
○ 637 papers in training set
Top 3%
4.8%
Similar papers in this journal
5
Molecular Systems Biology
○ 162 papers in training set
Top 0.4%
4.4%
Similar papers in this journal
50% of probability mass above
6
Nature Methods
○ 385 papers in training set
Top 3%
4.0%
7
Bioinformatics
◐ 1204 papers in training set
Top 5%
4.0%
8
Nature Biotechnology
○ 172 papers in training set
Top 2%
2.4%
9
Briefings in Bioinformatics
◐ 354 papers in training set
Top 3%
2.4%
10
Nature Machine Intelligence
○ 70 papers in training set
Top 1%
2.4%
11
PLOS Computational Biology
● 1863 papers in training set
Top 13%
1.9%
12
Genome Research
● 468 papers in training set
Top 4%
1.7%
13
Nature Biomedical Engineering
○ 47 papers in training set
Top 0.8%
1.7%
14
Proceedings of the National Academy of Sciences
● 2444 papers in training set
Top 31%
1.5%
15
npj Systems Biology and Applications
○ 125 papers in training set
Top 1%
1.4%
16
Advanced Science
○ 286 papers in training set
Top 6%
1.3%
17
Science Advances
● 1243 papers in training set
Top 25%
1.1%
18
Scientific Reports
○ 3612 papers in training set
Top 63%
1.1%
19
Nucleic Acids Research
◐ 1281 papers in training set
Top 11%
1.1%
20
eLife
● 5828 papers in training set
Top 59%
1.1%
21
iScience
○ 1154 papers in training set
Top 29%
1.1%
22
Genome Medicine
○ 183 papers in training set
Top 4%
1.1%
23
Nature Genetics
○ 286 papers in training set
Top 5%
1.0%
24
NAR Genomics and Bioinformatics
◐ 242 papers in training set
Top 4%
0.8%
25
PLOS ONE
● 5266 papers in training set
Top 62%
0.6%
26
Communications Chemistry
○ 48 papers in training set
Top 2%
0.6%
27
Communications Biology
○ 993 papers in training set
Top 35%
0.6%

"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.