MOD - A Machine-learning optimal detection method.

These functions show the essential steps of using MOD for detecting EPSP/EPSCs as described in [1] and applied in [2].

In order to run this demo, the following software packages are needed:

Copyright (C) 2016-2023 Alois Schlögl, IST Austria

Downloads: https://pub.ist.ac.at/~schloegl/software/mod/mod-2.3.tar.xz https://pub.ist.ac.at/~schloegl/software/mod/mod-2.3.zip

The source code is also available as part of Biosig [3], in subdirectory …/t340_MachineLearningOptimalDetection/

References:

[1] Zhang X, Schlögl A, Vandael D, Jonas P, MOD: A novel machine-learning optimal-filtering method for accurate and efficient detection of subthreshold synaptic events in vivo. Journal of Neuroscience Methods, 2021. https://doi.org/10.1016/j.jneumeth.2021.109125

[2] Xiaomin Zhang, Alois Schlögl, Peter Jonas, Selective Routing of Spatial Information Flow from Input to Output in Hippocampal Granule Cells, Neuron, 2020. https://doi.org/10.1016/j.neuron.2020.07.006 http://www.sciencedirect.com/science/article/pii/S0896627320305237

[3] Biosig - an Free and Open source software library for biomedical signal processing http://biosig.sourceforge.net/index.html

[4] The NaN-toolbox: A statistics and machine learning toolbox for Octave and Matlab® for data with and w/o MISSING VALUES encoded as NaN’s. https://pub.ist.ac.at/~schloegl/matlab/NaN/

[5] SigViewer (v0.5.1 or v0.65+) https://github.com/cbrnr/sigviewer/commit/f62f8d9d16db8c5c89c0fc12563b20bf4ad2ab7f (Nov 18, 2020) or later. Binaries for windows are available here: https://pub.ist.ac.at/~schloegl/software/sigviewer/sigviewer-20230620-win64.exe

[6] https://pub.ist.ac.at/~schloegl/software/mod/README.html

[7] https://en.wikipedia.org/wiki/Pattern_search_(optimization)

Content:

% demo - applying the method to the example data available from
%    https://pub.ist.ac.at/~schloegl/software/mod/
% results will be written into output/*
demo_mod.m
% An extended version with LOOM-based cross-validation, and two scorings
%   is shown in
demo_modx.m (obsolete, use demo_mody.m instead)

% improved version of demo_modx
demo_mody.m

% applying the general classifier to new data
apply_general_classifier.m

% loads the data, and apply preprocessing steps (High- and lowpass filtering, resampling to common sampling rate, AP detection and blanking, blanking of other artifacts like RS-pulses)
sload4mod.m

% this is the core function for obtaining the parameters of
    % from the MOD method  (filter coefficients, threshold, and delay)
mod_optimal_detection_filter.m

% storing the results in a GDF file for visualization with sigviewer
minidet2gdf.m

FAQ

How to apply it to your own data

This depends whether you want to build a detector for each cell, or whether you want to build a general classifier.

Cell-based classifier:

Check out demo_mod.m and address these items:

  1. you need your raw data, and your scorings and adapt loading the data (lines 67-87) Please note, the scorings consists of two periods (one segment in the beginning, the other from the end) And you need to tell which periods are scored, when constructing the scoring trace lines 124-136.

  2. In case you want to apply blanking of AP’s, you need to detect them and replace them with NaN’s (line 88-90, 115-123)

  3. if you want to downsample your data, you can check the lines 92-114,

  4. If you want to apply cross-validation, you should enable lines 140 - 194

That should do it. The main data processing of training the detection method happens in lines 199-213, the remainder storing the results, and in case you still have some HF noise in the raw detection trace, you might want to smooth it before doing the event detection.

General classifier:

Check out demo_mody and address these items:

  1. loading of data, where is your data located, how to load it. define the grouping information, this is useful for the cross-validation later

  2. get the scorings, identify which parts of the data is scored

  3. proprocessing, detection artifacts, AP etc. and remove (blank) those.

  4. run training and testing procedure on the scored data, use Leave-one-out-method for cross-validation.

  5. Build the general classifier from all scored data.

  6. (optional) check whether the changing default parameters for preprocessing (i.e. HP, LP, WINLEN, MAXLAG) would improve the result.

Is the output in the text file simply the event timings?

Yes, the event timing can be found in multiple outputs.

Changes

2023-06-27: v2.3

- restructuring of code, such that the codes for preprocessing, training,
  and applying the classifier is easier to identify. Therefore, demo_modx
  is now replaced with demo_mody, preprocessing is done in sload4mod, and
  the application of the GeneralClassifier is done in
  apply_general_classifier.

- sload4mod is a new fuction and provides additional pre-processing methods,
  including Highpass-filtering, Gaussian lowpass filter, and improves speed
  of downsampling

- apply_general_classifier: demonstrates how the precomputed classifier can
  be applied to new data.

- demo_mody has been used to optimize the preprocessing, using a manual
  patternsearch approach [7], suggesting that a different set of
  preprocessing would improve the results. The resulting changes are:
                             demo_modx        demo_mody
  HighPass filter:           detrend          1 Hz (Gaussian)
  LowPass filter;            1000 Hz          5000 Hz (Gaussian)
  WINLEN [ms]                4 ms             3 ms
  TemplateLength [samples]   1000             400

- get_local_maxima_above_threshold uses now another default method 
  using WINLEN/2 as a refractory period

2022-09-26: v2.2 - fix filenames in demo_mod.m

2022-05-13: v2.1 add smoothing of raw detection trace to demo enable cross-validation by default; report kappa in addition to AUC improvements and cleanup of demo_mod{,x}

2022-03-02: - demo_modx added - shows the use of leave-one-out-method for crossvalidation - add 6 data sets, with 2 scorings each - use accovf_mex to speed up training step

2021-06-21: improve documentation

2021-03-12: first release: