Back

Label-free multiplex microscopic imaging by image-to-image translation overcoming the trade-off between pixel- and image-level similarity

Morikura, T.; Funahashi, A.

2024-11-28 systems biology
10.1101/2024.11.25.625310 bioRxiv
Show abstract

Establishment of multiplex microscopic imaging without labeling is indispensable for understanding complex interactions of subcellular components. Toward the establishment of label-free multiplex microscopic imaging, image-to-image translation models that extract images of multiple subcellular components from bright-field images via nonlinear processing have attracted attention. However, the performance of conventional models is limited by a trade-off relationship between pixel- and image-level similarity, which degrades imaging performance. Here, we developed an image-to-image Wasserstein Schrodinger Bridge model to achieve high-performance image-to-image translation at the pixel level using Schrodinger Bridge while minimizing Wasserstein distance at the image level. Our model dramatically outperformed the conventional models at both levels simultaneously, reducing the mean squared error by 410-fold and improving the structural similarity index measure by 17.1-fold. Label-free multiplex microscopic imaging based on our model paves a way for the analysis of the interactions of subcellular components.

Matching journals

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

1
Nature Communications
○ 5641 papers in training set
Top 11%
15.3%
Similar papers in this journal
2
PLOS Computational Biology
● 1863 papers in training set
Top 2%
15.3%
Similar papers in this journal
3
Nature Methods
○ 385 papers in training set
Top 0.9%
10.8%
Similar papers in this journal
4
Nature Machine Intelligence
○ 70 papers in training set
Top 0.2%
8.0%
Similar papers in this journal
5
Bioinformatics
◐ 1204 papers in training set
Top 3%
6.8%
Similar papers in this journal
50% of probability mass above
6
Communications Biology
○ 993 papers in training set
Top 5%
3.3%
7
Patterns
○ 78 papers in training set
Top 0.7%
2.8%
8
Advanced Science
○ 286 papers in training set
Top 3%
2.8%
9
Briefings in Bioinformatics
◐ 354 papers in training set
Top 3%
2.5%
10
Genome Biology
○ 637 papers in training set
Top 4%
2.5%
11
eLife
● 5828 papers in training set
Top 41%
2.5%
12
iScience
○ 1154 papers in training set
Top 11%
2.4%
13
Scientific Reports
○ 3612 papers in training set
Top 48%
2.0%
14
Cell Reports Methods
○ 165 papers in training set
Top 1%
1.8%
15
Cell Systems
○ 201 papers in training set
Top 3%
1.1%
16
PLOS ONE
● 5266 papers in training set
Top 55%
1.1%
17
npj Systems Biology and Applications
○ 125 papers in training set
Top 2%
0.9%
18
Proceedings of the National Academy of Sciences
● 2444 papers in training set
Top 39%
0.9%
19
Nature Biotechnology
○ 172 papers in training set
Top 4%
0.9%
20
Science Advances
● 1243 papers in training set
Top 30%
0.9%
21
BMC Bioinformatics
○ 457 papers in training set
Top 6%
0.6%
22
Imaging Neuroscience
● 282 papers in training set
Top 4%
0.6%
23
npj Precision Oncology
○ 53 papers in training set
Top 2%
0.6%
24
IEEE Access
● 35 papers in training set
Top 1%
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.