Analyze dataset

The easiest way to use automea is to set up a csv file with a list of datasets and corresponding wells we want to analyze. The method analyze_dataset() takes as input a csv file and, based on parameters defined by the user, saves different output like statistics for each dataset, a list of spikes and bursts for each well, and more.

As an explample we have prepared the file ‘filenames_and_wells.csv’. Let’s load it with pandas to take a look at it.

import pandas as pd
csv_file = '../automea/final/filenames_and_wells_2.csv'
filenames_and_wells = pd.read_csv(csv_file, sep = ';')
filenames_and_wells

	filename	wells
0	AH22C001_3039_DIV14.h5	A6

The file contain the dataset named ‘AH22C001_3039_DIV14.h5’, and one well (A6).

import os
os.chdir('../automea/final/')

To perform the analysis we first need to define a set of parameters. Let’s start by importing automea and creating an object from the Analysis() class.

import automea

am = automea.Analysis()

One of the user defined parameters is which model one wants to use to perform the analysis - the machine learning models used for burst detection.

If the user wants to use one of the pretrained models available with the package, it’s only necessary to define the model_name. After this the method loadmodel() can be called, and the chosen model will be used for following analyses.

am.model_name = 'signal30.h5'
am.loadmodel()

We can check the model parameters, used for burst detection, by looking the at model_params attribute.

am.model_params

{'name': 'signal30.h5',
 'input_type': 'signal',
 'input_average': 30,
 'window_size': 50000,
 'window_overlap': 25000}

With the model loaded, we can define what we can to save from the analysis, by changing the analysis_params attribute.

am.analysis_params

{'save_spikes': False,
 'save_reverbs': False,
 'save_bursts': False,
 'save_net_reverbs': False,
 'save_net_bursts': False,
 'save_stats': False}

For instance, if we want to save only the high-level statistics, we set

am.analysis_params['save_stats'] = True

Now, we need to indicate where to find the csv file containing the datasets we want to analyze, the location of the datasets, and define a name for the output files.

am.path_to_csv = ''
csv_file = 'filenames_and_wells_2.csv'
am.path_to_dataset = '../../qneuron/mea-reproduce-results/mea-h5-datafiles/'
am.output_name = 'test'

Finally, we can run the analyze_dataset function for a csv file.

am.analyze_dataset(csv_file)

--- Running full analysis ---

--- Using model  signal30.h5  ---


 Analyzing dataset:  AH22C001_3039_DIV14.h5 

Well:  A6

--- Done! ---

While the code is running, a message shows which datasets and well are currently being analyzed.

After some minutes - or hours depending on how many datasets we want to analyze - a Done! message is shows indidcating that the process is finished.

The statistics we saved can be found in the file named ‘test_STATS_PREDICTED.csv’. We can import the file and look at the results.

stats = pd.read_csv('test_STATS_PREDICTED.csv')
stats

	Filename	Well Label	Number of channels	Total number of spikes	Mean Firing Rate [Hz]	Stray spikes (%)	Total number of networks bursts	Mean Network Bursting Rate [bursts/minute]	Mean Network Burst Duration [ms]	NIBI	CV of NIBI	Mean reverb per burst	Median of reverb per burst	Mean net reverb per net burst	Median of net reverb per net burst	Total number of network reverb	Mean net reverb frequency [reverb/min]	Mean net reverb duration [ms]	Mean in-reverb freq [Hz]
0	AH22C001_3039_DIV14.h5	A6	12	56193	7.8	11.92	49	4.9	759.41	11199.72	0.27	4.84	5.0	4.73	5.0	237	23.7	91.77	189.31

The file contains statistics about the dataset/well analyzed.