Tag Archives: Bayesian

Bayesian machine learning analysis of single-molecule fluorescence colocalization images

From Science at Brandeis: “Yerdos Ordabayev et al. in the Department of Biochemistry use Bayesian probabilistic programming to implement computer software “Tapqir” for analysis of colocalization single-molecule spectroscopy (CoSMoS) image data. CoSMoS is a tool widely used in vitro to study the biochemical and physical mechanisms of the protein and nucleic acid macromolecular “machines” that perform essential biological functions. In this method, formation and/or dissociation of molecular complexes is observed by single-molecule fluorescence microscopy as the colocalization of binder and target macromolecules each labeled with a different color of fluorescent dye. Despite the use of the method for over twenty years, reliable analysis of CoSMoS data remains a significant challenge to the effective and more widespread use of the technique.

This work describes a holistic causal probabilistic model of CoSMoS image data formation. This model is physics-based and includes realistic shot noise in fluorescent spots, camera noise, the size and shape of spots, and the presence of both specific and nonspecific binder molecules in the images. Most importantly, instead of yielding a binary spot-/no-spot determination, the algorithm calculates the probability of a colocalization event. Unlike alternative approaches, Tapqir does not require subjective threshold settings of parameters so they can be used effectively and accurately by non-expert analysts. The program is implemented in the state-of-the-art Python-based probabilistic programming language Pyro (open-sourced by Uber AI Labs in 2017), which enables efficient use of graphics processing unit (GPU)-based hardware for rapid parallel processing of data and facilitates future modifications to the model. Tapqir is free, open-source software. We envision that [the] program is likely to be adopted by researchers who use single-molecule colocalization methods to study a wide range of different biological systems.”

Yerdos is a postdoctoral fellow jointly advised by Profs. Douglas Theobald and Jeff Gelles.

 

10.7554/eLife.73860
Ordabayev Y.A., et al. Bayesian machine learning analysis of single-molecule fluorescence colocalization images
eLife, 11, e73860 (2022)