spotfitlm 0.1.3

MLE 2D gaussian fitting with Poisson deviates

spotfitlm

A small Python library for doing robust spot detection in 2D by (MLE) Gaussian fitting. The fitting is done through a small C library implementing a damped Gauss-Newton optimization algorithm (like Levenberg-Marquardt) algorithm. Error estimates (covariance) matrix of the fit parameters are computed from the (inverse of) full Hessian matrix. Currently, only a symmetric Gaussian fit is implemented.

The objective function being minimized is the same as Laurence & Chromy's MLE method [https://www.nature.com/articles/nmeth0510-338].

For spot detection, this package uses the algorithm from Danuser Lab's U-track MATLAB software. Specifically, the code from: [https://github.com/DanuserLab/u-track3D/blob/9279b3784de64d29bb06c3693e99f2e5c064288e/software/pointSourceDetection.m#L111]

for doing the hypothesis testing on significant Gaussian peaks above background noise.

Installation

This package is not yet released at PyPI so it can't be installed by a simple pip install, so you'll need to use git to install it to your Python environment:

For MacOS or Linux system make sure you have a c compiler accessible in your path. Windows installation is more complicated because I haven't figured out how to use the right compilers.

1. Clone this repository

git clone https://github.com/delnatan/spotfitlm.git

2. and go into the directory spotfitlm and run:

pip install -e .

Windows installation

Unfortunately, I haven't had the time to work out installation procedure for Windows. You just need to compile the C library and put it in a place where the .dll file can be loaded via Python. In the past, I setup the compiler within a conda environment, and installed m2w64-toolchain via conda-forge channels. Then you'll have access to gcc. Since there are no dependencies, the library should compile fine. I intend to make a simple Makefile for doing this process to streamline the setup in Windows machine.