G5M uses modified Gaussian mixture modeling to accurately capture the shape of 2D and 3D astigmatic localization clouds. In realistic simulations, it resolves molecules spaced 2.9 localization precisions apart with a 27× higher recovery rate than other tools. 3/7
Bonus: we found a new formula for axial loc. precision for astigmatic imaging for any z position. It only depends on the 3D calibration and the properties of localizations. Interestingly, the results are not necessarily symmetric around z = 0. 5/7
In DNA-PAINT, nearby proteins create overlapping localizations. You may see two by eye, but analysis often misses them, wasting precious precision. G5M solves this. Now in @natcomms.nature.com. doi.org/10.1038/s414.... 1/7
We validated G5M on DNA origami, Nuclear Pore Complexes (NPCs) and Obinutuzumab-treated CD20 RESI data. In NPCs, G5M found twice as many Nup96 dimers as GA, demonstrating increased accuracy in molecular mapping. 4/7
G5M is implemented in Picasso since version 0.9.5: github.com/jungmannlab/..., although we always recommend using the most recent version! A short documentation: picassosr.readthedocs.io/en/latest/re.... We advise reading the whole paper for best results! 6/7
We hope G5M pushes molecular mapping forward. Excited to see the discoveries it enables! Thanks to @rafalkowalew.bsky.social, @susannereinhardt.bsky.social, @ipachmayr.bsky.social, Shuhan Xu, @lumasullo.bsky.social for developing G5M and the @jungmannlab.bsky.social for testing and improving it. 7/7
Knowing where biomolecules are is key to uncovering molecular interactions and patterns. But standard clustering methods (e.g., DBSCAN, GA) often fail when proteins are spaced <5× the localization precision, leading to false negatives and misassignments. 2/7