A Bayesian model for single-cell gene expression analysis with MERFISH data

Johannes Köster

 

2016

www.medicalxpress.com

Genes

biosocialmethods.isr.umich.edu

From genes to proteins

Gene activity/expression:

the more RNA, the more protein

FISH

Fluorescence in-situ hybridization:

• label RNA with fluorescent probes
• see RNA molecules in single cells

 

Problem:

needs 1 color per gene

MERFISH

...

Chen et al. 2015

MERFISH

Problem with raw counts:

up to 20% loss and misidentification

Research question

Can we recover the true counts from the raw, biased MERFISH counts?

Approach

Event probabilities

Known error probabilities:

1→0 error: 1% - 10%

0→1 error: 0.5% - 4%

Urn model for scenario probabilities

Scenario:

one combination of events that explains observed counts given true counts

Likelihood of observed counts given truth

Bayesian model for differential gene activity

Results

Simulation

Pr(0→1)    Pr(1→0)

simulated

hybridization

Bayesian model recovers biased counts

Bayesian model recovers biased counts

Conclusion

Conclusion

• MERFISH is a technology to visualize and count RNA molecules in single cells.
• Raw counts are biased.

 

The presented Bayesian model provides:

• accurate estimates of the true RNA counts
• a way to access differential gene activity between cells while maintaining uncertainty information
• ability to measure more genes in parallel

Acknowledgements

Shirley Liu

Myles Brown

Bo Li

Peng Jiang

Eric Severson