LiteLoc

Scalable and lightweight deep learning for efficient high accuracy single-molecule localization microscopy

LiteLoc is a Python and Pytorch based scalable and lightweight deep learning for efficient high accuracy single-molecule localization microscopy (SMLM).
LiteLoc includes a scalable and competitive data analysis framework and a lightweight deep learning network which has small number of parameters (only 1.33 M) and low computational complexity (71.08 GFLOPs, 10 images with 128*128 pixels) while maintaining comparable or even better localization precision.
With the help of high parallelism between the data loader/analyzer/saver processes, the total analysis speed is ~25 times faster than that of DECODE and more than 560 MB/s data analysis throughput could be achieved with eight NVIDIA GTX 4090 GPUs.

System Requirements

OS: Linux(GPU accelerated) / Windows(GPU accelerated)
CUDA version: 12.1
Software: conda, anaconda

Python Environment Configuration

1. create a virtual environment using conda

Note: Generally python=3.8 is okay, but should be lower than 3.9.

conda create -n liteloc_env python==3.8.19

2. activate liteloc environment:

conda activate liteloc_env

3. install packages imported in liteloc

Note: If you cuda version is not 12.1 (Linux), please download the version of torch from Pytorch according to your operation system and cuda version.

pip install -r requirements.txt

conda install -c turagalab -c conda-forge spline

Quick Start! (Demo of Figure 3a and Figure 3d in LiteLoc paper)

Demo of Figure 3a: train a network based on our experimental astigmatism NPC data.
Demo of Figure 3d: train a network based on our experimental 6μm DMO-tetrapod NPC data.
Download link (PSF model and data):

1. get PSF model and experimental data.

We also provide GUI in Matlab to generate your own PSF model. You can run /PSF Modeling/Fit_PSF_model/calibrate_psf_model_GUI.m program, load your own experimental beads .tif file, set parameters and finally get a .mat file, which includes both vectorial and cspline PSF model. This file will be loaded automatically in the training process to generate training data.

2. train your own LiteLoc

Please uncompress the downloaded data in step 1, then put astigmatism NPC data into the directory /datasets/demo-fig3a/, 6μm DMO-tetrapod NPC into /datasets/demo-fig3d/, put astigmatic PSF model into the directory /calibrate_mat/demo-fig3a/ and 6μm tetrapod PSF into /calibrate_mat/demo-fig3d/.
Then check the 'calibration_path', 'result_path' and 'infer_data' in 'train_params_demo_fig3a/3d.yaml'. Please pay attention to the parameters with notes.
Note: If you run the program in terminal directly, please close the figure window that program plots at the beginning to continue the training.

3. infer your data and get results

Similar to the training setup, you need to set parameters in the template file for inference according your computation resource.