NN diffusivity mapping

Code repository for Pix2D diffusivity mapping. Requires matlab-storm published by Zhuanglab (https://github.com/ZhuangLab/matlab-storm)
Codes are written to process images in .dax format and localization data from Insight3. Localization with Insight3 can be replaced with Python: https://github.com/ZhuangLab/storm-analysis

Training data generation

• Use codes under Matlab > Simulation
• Training sets with multiple single-molecule brightness and shot noise level can be generated with Batch_Training_set_generation.m
• Set exposure time, pixel length, gain factor and PSF size to match your imaging setup.
• Run localization analysis to generated Dax image files.
• Generate training input from .dax and .bin file with Training_set_extraction.m
• Generate augmented training input with training_set_combine.m

Training CNN

• Use codes under Matlab > Training
• To sweep hyperparamters, run Batch_training_parameter_sweep.m
• Run Batch_training.m for to get trained CNN

Processing Dax image files

• Run localization for single-molecule images
• Use Data sampling > Batch_ROI_extraction to extract single-molecule signals and positions
• Run Data processing > Batch_D_mapper_from_array.m to generate spatial diffusivity map in desired bin size
• Run Data processing > Dmap_rendering.m to plot 2D diffusivity map in color scale

Contact: Ha Park (ha.park@berkeley.edu)