HD-MEA NEUROPulse

Description

Abstract (English)

The analysis of data for scientific research often requires routines that go beyond the standard preprocessing provided by companies along with the hardware for signal recording. We present a Python-based script for data extraction from 3Brain© HD-MEA datasets recorded using Brainwave© X or 4, 5 or 6. The code is optimized for cerebellar slice recordings (including both evoked and spontaneous activities) but it can be adapted to various types of recordings with minor adjustments. This tool is designed to facilitate data extraction and visualization from 3Brain© systems. With some modifications, it can also be applied to other multi-electrode array systems. The program efficiently extracts information directly from the raw Brainwave file (.brw) and processed datasets (.bxr, with spike detection and sorting results). The program output consists of .csv files with the extracted data and corresponding plots. The GUI structure is straightforward, featuring the following panels:

• manual selection of channels to be analyzed
• file paths specification
• saving folders definition
• region-of-interest (ROI) or channel group (.bwcg) files.

Three distinct panels cope with different types of analysis. The first panel is dedicated to evoked local field potential (LFP) detection, parameter extraction, and result plotting. The second panel focuses on autorhythmic unit evoked activity and generates peri-stimulus time histograms (PSTHs). The third panel manages spontaneous activity, generating inter-spike interval (ISI) plots and calculating the average firing rate ± SEM (standard error of the mean), modal firing rate (which could be helpful for long recordings), and firing rate variability (CV and CV2¹).

Methods

INSTRUCTIONS

The program is available as a .exe file, which can be run by double-clicking the icon (ensure not to delete the auxiliary files in the folder). The Python source code is also provided, but specific libraries are needed to run it.

Additionally, we provide a dataset (including both .brw and .bxr files) with a pre-selected ROI (.bwcg) for both LFP and autorhythmic cell analysis.

For Autorhythmic Cells Evoked Activity Analysis (PSTH) and Autorhythmic Cells Spontaneous Activity Analysis (ISI), spike detection and spike sorting are required: these steps can be easily performed directly in Brainwave© and exported as a .bxr file. Notice that sorting should be supervised in order to have one unit per channel.

File Requirements for Different Analyses:

• LFP Analysis: Requires only the .brw file.
• PSTH Analysis: Requires both .brw and .bxr files.
• ISI Analysis: Requires only the .bxr file.          

GENERAL OPERATIONS

As described above, the first panel of the GUI, named ID Conversion, is dedicated to the manual selection of ROIs (in contrast to ROI upload). In this panel, you can input the coordinates of the channels to be analysed, one by one. Next to this panel, a second panel, named File Path, allows to upload information from files. It contains four text fields allowing to specify the paths for the .brw file, .bxr file, saving folder, and ROI file. For convenience, you can also select these paths by browsing the folders using the white folder icon next to each text field. Each manually added channel or loaded ROI is displayed in green on the MEA virtual map (the blue square at the top right corner), demonstrating that the ROI is correctly loaded. Once all files are correctly loaded and the ROI is selected, the analysis can begin. Regardless of the type of analysis, the data is initially stored as digital values in a machine-friendly format. The program's first operation is to convert this digital data into a format that can be easily visualized and analysed. Once the data is reorganized, the actual analysis can be performed.

LFP ANALYSIS

The field potential analysis is fully optimized for cerebellar granular layer evoked field potentials, whose waveform is well characterized ^2,3,4. The LFP Analysis panel features four parameters:

• LFP Extraction Time: The total duration of the trace to be extracted and plotted (consider that the program uses the entire trace for statistical validation of the signal and that this time needs to be refined depending on the stimulation frequency).
• N_2a N_2b Time Window: The time interval between the N_2a and N_2b peaks, which are characteristic negative peaks of cerebellar granular layer LFP.
• Blind Time: The time window that should be excluded as the final part of the stimulus artifact.
• Low Pass Filter Frequency: A low-pass filter where you can select the cutoff frequency.

The LFP analysis algorithm begins by detecting every stimulation frame and storing it in an array. The second part of the script detects the first minimum (the N_2a peak) and the second minimum (the N_2b peak) in the trace extracted between the stimulus frame plus the blind time and the  last frame in the ‘LFP extraction time’. Results are stored in an array. These operations are repeated for each field potential in each channel within the ROI. At the end, results from each channel ID, including the average trace, are grouped, plotted and stored in a matrix. The program performs a statistical validation to select only signals that exceed the noise. Since data in the extracted trace follow a normal distribution, the required peak is the outlier. Based on this idea, the program sets a threshold for each LFP in each channel equal to MEAN - 3*STDEV. This threshold ensures 95% confidence for the detected peak. The final part of the routine is dedicated to saving results, which are exported as a .csv file containing peak amplitudes and a .csv file with time frames of stimulus delivery. Additionally, every LFP group is plotted and saved as a .png file (it can also be saved as a vector by changing .png to .svg) for each channel.

ANALYSIS OF EVOKED ACTIVITY IN AUTORHYTHMIC CELLS

The Autorhythmic Cells Evoked Activity Analysis (PSTH) analysis panel allows to set six parameters:

• Bin Size: The size of the bin used to count spikes and calculate the PSTH bin frequency.
• Pre-Stimulus Time Window: The desired time interval to be analysed before stimulus delivery.
• Post-Stimulus Time Window: The desired time interval to be analysed after stimulus delivery .
• Number of Permutations for Significance Test: The number of random permutations performed during the permutation test for statistical significance between pre- and post-stimulus groups.
• Frequency of Stimulation Train Delivered.
• Number of Stimulation Pulses Delivered.

The PSTH analysis algorithm starts with stimulus detection, similarly to the LFP analysis. Then, it builds an array for every channel in the ROI and for each bin in the selected time windows (merging the previously defined pre- and post-stimulus time windows). It then counts the number of events (spikes) in each bin for every delivered stimulus, divides by the number of delivered stimuli, and calculates the frequency in Hz for data visualization. At this point, the algorithm removes the stimulus artifact tag (sometimes the stimulus artifact is mistakenly considered a spike) and performs two kinds of statistical validation. The first is a permutation test between the pre-stimulus and the post-stimulus groups. The second is a z-score statistic comparing the pre-stimulus group to each bin of the post-stimulus group. The algorithm generates a PSTH for each channel, marking significant bins with a red star. The output of the algorithm includes a .csv file with the results, the PSTHs for each channel as .png files, and a summary plot of all channels activity in colorscale as a .png file.

*Z = (post stim freq bin - pre stim freq mean) / pre stim freq stdev           Z>1.645 or Z<-1.645 for 95% significance

SPONTANEOUS ACTIVITY ANALYSIS

The Autorhythmic Cells Spontaneous Activity Analysis (ISI) panel allows to set three parameters:

• Bin Size: The bin size used for ISI calculation and plotting.
• ISI Bin Limit: The maximum ISI considered.
• Total Time: The full length of the recordings.

The spontaneous activity analysis algorithm calculates the ISI for each channel in the ROI, the average firing rate ± SEM, the modal firing rate, CV and CV2. The outputs of the algorithm are .csv files containing both the ISI data and the features of each unit, and ISI plots for each channel as .png files. 

CV = stdev(ISI) / mean firing rate (ISI)                        CV2 = 2 |ISIn+1 - ISIn| / |ISIn1+ISIn|  

Technical info

Files in the repository should be grouped in two folders

• NEUROPULSE: array.dat, channels ids table.csv, folderico.png, HD-MEA NEUROPulse.exe, HD-MEA NEUROPulse.py, img.png, img_map.png, NeuroPulseLogo.png, python.png
• DATASET: example.brw, example.bxr, example_SA.bxr, ROI_LFP.bwcg, ROI_PCs.bwcg, ROI_PCs_stim.bwcg

follow instruction in EXAMPLE ANALYSIS PIPELINE.pdf to perform data analysis on given example files.