research
Computational imaging and discovery, from molecules to cells.
From noisy images to molecular discovery
We combine physics-based modeling and AI to detect molecules and characterize their structural states directly from cellular cryo-EM/ET data. Our ultimate goal is to understand how the structures of molecular machines relate to their functions in cells, and to use these insights to guide the design of more effective therapeutics.
01 DiffPose
Faster in situ template searches
We speed up projection matching so cellular cryo-EM images can be searched against many molecular templates. Our goal is to make large-scale in situ searches practical for identifying molecules directly inside cells.
02 GisAPR
Refining atomic models directly from images
We adjust the atomic model to match observed 2D cryo-EM images directly. This lets us refine molecular structures without first reconstructing a 3D map.
03 Sub-50 kDa
Resolving structures below 50 kDa
We use known structures as templates to select and align images of proteins smaller than 50 kDa. This improves single-particle cryo-EM reconstructions, revealing clearer details of ligands and their binding pockets.
04 2DTM p-value
Detecting small and rare molecular targets
We developed the 2DTM p-value, a statistical score that helps distinguish molecular signals from noise in cryo-EM images. It improves detection of small and rare targets while controlling false positives.
05 In situ classification
Distinguishing molecular states inside cells
We distinguish different stages of ribosome assembly directly in cellular cryo-EM images, without reconstructing a 3D map. Using multiple templates and a statistical model, we classify individual ribosomes and map where these distinct states occur inside cells.
These projects include current work and earlier research by Kexin Zhang with collaborators. Explore each project for more figures, animations, and publications.