Deep learning for pixelwise labeling of 3d biological data

Helices and beta sheets are characterized by cylindrical and planar shapes respectively

Cryo-Em allows us to see these structures at a variety of resolution

At medium resolution we can barely see the structure

After attempting multiple architectures, a custom 3d Unet provided the best results

Results

Simulted

Experimental

Authors

• Devin Haslam - All programing and research