Configuring Metadata¶
To load your data with neurotic, you must organize them in one or more YAML files, called metadata files.
YAML files are very sensitive to punctuation and indentation, so mind those details carefully! Importantly, the tab character cannot be used for indentation; use spaces instead. There are many free websites that can validate YAML for you.
You may include comments in your metadata file, which should begin with #
.
Top-Level Organization¶
Datasets listed within the same metadata file must be given unique names, which
may include spaces. The special name neurotic_config
is reserved for
neurotic configuration settings and cannot be used for datasets.
In addition to names, a long description can be provided for each dataset.
Details pertaining to each dataset, including the description, are nested
beneath the dataset name using indentation. You may need to use double quotes
around names, descriptions, or other text if they contain special characters
(such as :
or #
) or are composed only of numbers (such as a date).
experiment 2020-01-01:
description: Both the name and description will be visible when neurotic loads the metadata
# other details about this dataset will go here
my favorite dataset:
description: This time it actually worked!
# other details about this dataset will go here
Specifying Data Locations¶
Within a dataset’s YAML block, paths to data and video files should be provided.
All files associated with a dataset should be collected into a single
directory. A path to the local copy of this directory can be provided using the
data_dir
key. You may specify data_dir
as an absolute path (e.g.,
C:\Users\me\folder
) or as a path relative to the metadata file (e.g.,
folder
). If left unspecified, the directory containing the metadata file is
used.
Paths to individual files within the dataset are provided using keys listed
below. These paths should be given relative to data_dir
. If data_dir
is
flat (no subdirectories), these should be simply the file names.
Key | Description |
---|---|
data_file |
A single Neo-compatible data file (see neo.io
for file formats) |
video_file |
A video file that can be synchronized with data_file |
annotations_file |
A CSV file for read-only annotations |
epoch_encoder_file |
A CSV file for annotations writable by the epoch encoder |
tridesclous_file |
A CSV file output by tridesclous’s DataIO.export_spikes |
Note that the annotations_file
must contain exactly 4 columns with
these headers: “Start (s)”, “End (s)”, “Type”, and “Label”.
The epoch_encoder_file
must contain exactly 3 columns with these headers:
“Start (s)”, “End (s)”, and “Type”. (The fourth column is missing because
ephyviewer’s epoch encoder is currently unable to attach notes to individual
epochs; this may be improved upon in the future.)
The tridesclous_file
is described in more detail in
tridesclous Spike Sorting Results.
Remote Data Available for Download¶
Data files must be stored on the local computer for neurotic to load them
and display their contents. If the files are available for download from a
remote server, neurotic can be configured to download them for you to the
local directory specified by data_dir
if the files aren’t there already.
Specify the URL to the directory containing the data on the remote server using
remote_data_dir
. neurotic expects the local data_dir
and the
remote_data_dir
to have the same structure and will mirror the
remote_data_dir
in the local data_dir
when you download data (not a
complete mirror, just the specified files).
For an example, consider the following:
my favorite dataset:
data_dir: C:\Users\me\folder
remote_data_dir: http://myserver/remote_folder
data_file: data.axgx
video_file: video.mp4
With a metadata file like this, the file paths data_file
and video_file
are appended to remote_data_dir
to obtain the complete URLs for downloading
these files, and they will be saved to the local data_dir
.
If you have many datasets hosted by the same server, you can specify the server
URL just once using the special remote_data_root
key, which should be
nested under the reserved name neurotic_config
outside of any dataset’s
YAML block. This allows you to provide for each dataset a partial URL to a
folder in remote_data_dir
which is relative to remote_data_root
. For
example:
neurotic_config: # reserved name for global settings
remote_data_root: http://myserver
my favorite dataset:
data_dir: C:\Users\me\folder1
remote_data_dir: remote_folder1
data_file: data.axgx
video_file: video.mp4
another dataset:
data_dir: C:\Users\me\folder2
remote_data_dir: remote_folder2
data_file: data.axgx
video_file: video.mp4
Here, URLs to video files are composed by joining remote_data_root
+
remote_data_dir
+ video_file
.
Recall that if data_dir
is a relative path, it is assumed to be relative
to the metadata file. In the example above, if the metadata file is located in
C:\Users\me
, the paths could be abbreviated:
neurotic_config:
remote_data_root: http://myserver
my favorite dataset:
data_dir: folder1
remote_data_dir: remote_folder1
data_file: data.axgx
video_file: video.mp4
another dataset:
data_dir: folder2
remote_data_dir: remote_folder2
data_file: data.axgx
video_file: video.mp4
Note
Portability is easy with neurotic! Use relative paths in your metadata file along with a remotely accessible data store such as GIN to make your metadata file fully portable. The example above is a simple model of this style. A metadata file like this can be copied to a different computer, and downloaded files will automatically be saved to the right place. Data stores can be password protected and neurotic will prompt you for a user name and password. This makes it easy to share the neurotic experience with your colleagues! 🤪
URLs to Use with GIN¶
If you have data stored in a public repository on GIN, you can access it from a URL of this form:
https://gin.g-node.org/<username>/<reponame>/raw/master/<path>
For private repositories, you must use a different URL that takes advantage of the WebDAV protocol:
https://gin.g-node.org/<username>/<reponame>/_dav/<path>
The second form works with public repos too, but GIN login credentials are still required. Consequently, the first form is more convenient for public repos.
Global Configuration Settings¶
The top-level name neurotic_config
is reserved for configuration settings
that apply to all datasets or to the app itself. The following settings may be
nested beneath neurotic_config
.
Key | Description |
---|---|
neurotic_version |
A version specification stating the version of neurotic required by the metadata. Presently, if the requirement is not met, only a warning is issued. Quotation marks around the spec are usually required. |
remote_data_root |
A URL prepended to each remote_data_dir that is not
already a full URL (i.e., does not already begin with a
protocol scheme like https:// ) |
For example:
neurotic_config:
neurotic_version: '>=1.4,<2'
remote_data_root: http://myserver
my favorite dataset:
# dataset details here
Data Reader (Neo) Settings¶
The electrophysiology file specified by data_file
is read using Neo, which
supports many file types. A complete list of the implemented formats can be
found here: neo.io
.
By default, neurotic will use the file extension of data_file
to guess
the file format and choose the appropriate Neo IO class for reading it. If the
guess fails, you can force neurotic to use a different class by specifying
the class name with the io_class
parameter (all available classes are
listed here: neo.io
).
Some Neo IO classes accept additional arguments beyond just a filename (see the
Neo docs for details: neo.io
). You can specify these arguments in your
metadata using the io_args
parameter.
For example, suppose you have data stored in a plain text file that is missing
a file extension. The neo.io.AsciiSignalIO
class can read plain text
files, but you must specify this manually using io_class
because the
extension is missing. You could do this and pass in supported arguments in the
following way:
my favorite dataset:
data_file: plain_text_file_without_file_extension
io_class: AsciiSignalIO
io_args:
skiprows: 1 # skip header
delimiter: ' ' # space-delimited
t_start: 5 # sec
sampling_rate: 1000 # Hz
units: mV
Video Synchronization Parameters¶
Constant Offset¶
If data acquisition began with some delay after video capture began, provide a
negative value for video_offset
equal to the delay in seconds. If video
capture began after the start of data acquisition, use a positive value. A
value of zero will have no effect.
neurotic warns users about the risk of async if video_file
is given but
video_offset
is not. To eliminate this warning for videos that have no
delay, provide zero.
Frame Rate Correction¶
If the average frame rate reported by the video file is a little fast or slow,
you may notice your video and data going out of sync late in a long experiment.
You can provide the video_rate_correction
parameter to fix this. The
reported average frame rate of the video file will be multiplied by this factor
to obtain a new frame rate used for playback. A value less than 1 will decrease
the frame rate and shift video events to later times. A value greater than 1
will increase the frame rate and shift video events to earlier times. A value
of 1 has no effect.
You can obtain a good estimate of what value to use by taking the amount of time between two events in the video and dividing by the amount of time between the same two events according to the data record (seen, for example, as synchronization pulses or as movement artifacts).
Discrete Desynchronization Events¶
If you paused data acquisition during your experiment while video capture was
continuous, you can use the video_jumps
parameter to correct for these
discrete desynchronization events, assuming you have some means of
reconstructing the timing. For each pause, provide an ordered pair of numbers
in seconds: The first is the time according to data acquisition (not
according to the video) when the pause occurred, and the second is the duration
of the pause during which the video kept rolling.
For example:
my favorite dataset:
video_file: video.mp4
# etc
video_jumps:
# a list of ordered pairs containing:
# (1) time in seconds when paused occurred according to DAQ
# (2) duration of pause in seconds
- [60, 10]
- [120, 10]
- [240, 10]
These values could correct for three 10-second pauses occurring at times 1:00, 2:00, 3:00 according to the DAQ, which would correspond to times 1:00, 2:10, 3:20 according to the video. The extra video frames captured during the pauses will be excised from playback so that the data and video remain synced.
neurotic will automatically suggest values for video_jumps
if it reads
an AxoGraph file that contains stops and restarts (only if video_jumps
is
not already specified).
Real-World Date and Time¶
The GUI can optionally display the real-world date and time. This feature is accurate only if the recording is continuous (no interruptions or pauses during recording) and the start time of the recording is known. Some data file formats may store the start time of the recording, in which case neurotic will use that information automatically. However, if the start time is missing or inaccurate, it can be specified in the metadata like this:
my favorite dataset:
data_file: data.axgx
rec_datetime: 2020-01-01 13:14:15
# etc
Plot Parameters¶
Use the plots
parameter to specify which signal channels from data_file
you want plotted and how to scale them. Optionally, a color may be specified
for channels using a single letter color code (e.g., 'b'
for blue or
'k'
for black) or a hexadecimal color code (e.g., '1b9e77'
).
Consider the following example, and notice the use of hyphens and indentation for each channel.
my favorite dataset:
data_file: data.axgx
# etc
plots:
- channel: Extracellular
ylabel: Buccal nerve 2 (BN2)
units: uV
ylim: [-150, 150]
color: r
- channel: Intracellular
ylabel: B3 neuron
units: mV
ylim: [-100, 50]
color: '666666'
- channel: Force
units: mN
ylim: [-10, 500]
This would plot the “Extracellular”, “Intracellular”, and “Force” channels from
the data_file
in the given order. ylabel
is used to relabel a channel
and is optional. The units
and ylim
parameters are used together to
scale each signal such that the given range fits neatly between the traces
above and below it. If units
is not given, they are assumed to be
microvolts for voltage signals and millinewtons for force signals. If ylim
is not given, they default to [-120, 120]
for voltages and [-10, 300]
for forces.
If plots
is not provided, all channels are plotted using the default
ranges, except for channels that match these patterns: “Analog Input #*” and
“Clock”. Channels with these names can be plotted if given explicitly by
plots
.
Time Range¶
The amount of time initially visible can be specified in seconds with
t_width
.
The position of the vertical line, which represents the current time in each
plot, can be specified as a fraction of the plot range with past_fraction
.
A value of 0 places the vertical line at the left edge of each plot;
consequently, everything plotted is “in the future”, occurring after the
current time. A value of 1 places the vertical line at the right edge of each
plot; consequently, everything plotted is “in the past”, coming before the
current time. The default value of 0.3 causes the first 30% of the plot range
to display “the past” and the last 70% to display “the future”.
Epoch Encoder Parameters¶
The labels available to the epoch encoder must be specified ahead of time using
epoch_encoder_possible_labels
(this is a current limitation of ephyviewer
that may eventually be improved upon).
For example:
my favorite dataset:
epoch_encoder_file: epoch-encoder.csv
# etc
epoch_encoder_possible_labels:
- label1
- label2
- label3
Filters¶
Highpass, lowpass, and bandpass filtering can be applied to signals using the
filters
parameter. Note that filters are only applied if fast loading is
off (lazy=False
).
Consider the following example, and notice the use of hyphens and indentation for each filter.
my favorite dataset:
data_file: data.axgx
# etc
filters: # used only if fast loading is off (lazy=False)
- channel: Extracellular
highpass: 300 # Hz
lowpass: 500 # Hz
- channel: Intracellular
highpass: 300 # Hz
- channel: Force
lowpass: 50 # Hz
Filter cutoffs are given in hertz. Combining highpass
and lowpass
provides bandpass filtering.
Amplitude Discriminators¶
Spikes with peaks that fall within amplitude windows given by
amplitude_discriminators
can be automatically detected by neurotic on
the basis of amplitude alone. Note that amplitude discriminators are only
applied if fast loading is off (lazy=False
).
Detected spikes are indicated on the signals with markers, and spike trains are
displayed in a raster plot. Optionally, a color may be specified for an
amplitude discriminator using a single letter color code (e.g., 'b'
for
blue or 'k'
for black) or a hexadecimal color code (e.g., '1b9e77'
).
In addition to restricting spike detection for a given unit to an amplitude window, detection can also be limited in time to overlap with epochs with a given label.
Consider the following example, and notice the use of hyphens and indentation for each amplitude discriminator.
my favorite dataset:
data_file: data.axgx
# etc
amplitude_discriminators: # used only if fast loading is off (lazy=False)
- name: Unit 1
channel: Extracellular
units: uV
amplitude: [50, 150]
color: r
- name: Unit 2
channel: Extracellular
units: uV
amplitude: [20, 50]
epoch: Unit 2 activity
color: 'e6ab02'
Here two units are detected on the same channel with different amplitude
windows. Any peaks between 50 and 150 microvolts on the “Extracellular” channel
will be tagged as a spike belonging to “Unit 1”. The discriminator for “Unit 2”
provides the optional epoch
parameter. This restricts detection of “Unit 2”
to spikes within the amplitude window that occur at the same time as epochs
labeled “Unit 2 activity”. These epochs can be created by the epoch encoder
(reload required to rerun spike detection at launch-time), specified in the
read-only annotations_file
, or even be contained in the data_file
if
the format supports epochs.
Amplitude windows are permitted to be negative.
tridesclous Spike Sorting Results¶
tridesclous is a sophisticated spike sorting toolkit. The results of a sorting
process can be exported to a CSV file using tridesclous’s
DataIO.export_spikes
function. This file contains two columns: the first is the sample index of a
spike, and the second is the ID for a cluster of spikes. If this file is
specified with tridesclous_file
, then a mapping from the cluster IDs to
channels must be provided with tridesclous_channels
.
In the following example, notice the lack of hyphens:
my favorite dataset:
data_file: data.axgx
tridesclous_file: spikes.csv
# etc
tridesclous_channels:
0: [Channel A, Channel B]
1: [Channel A]
2: [Channel B]
3: [Channel B]
# etc
Here numeric cluster IDs are paired with a list of channels found in
data_file
on which the spikes were detected.
To show only a subset of clusters or to merge clusters, add the
tridesclous_merge
parameter.
In this example, note again the punctuation:
my favorite dataset:
data_file: data.axgx
tridesclous_file: spikes.csv
# etc
tridesclous_channels:
0: [Channel A, Channel B]
1: [Channel A]
2: [Channel B]
3: [Channel B]
# etc
tridesclous_merge:
- [0, 1]
- [3]
Now clusters 0 and 1 are combined into a single unit, and only that unit and cluster 3 are plotted; cluster 2 has been discarded.
Firing Rates¶
If spike trains were generated using
Amplitude Discriminators, imported from
tridesclous Spike Sorting Results, or included in the data_file
, their
smoothed firing rates can be computed. Note that firing rates are computed only
if fast loading is off (lazy=False
).
Firing rates are plotted as continuous signals. Colors are inherited from
amplitude_discriminators
, if they are provided there.
Firing rates are computed using a kernel that is convolved with the spike train. The metadata is specified like this:
my favorite dataset:
data_file: data.axgx
# etc
amplitude_discriminators: # used only if fast loading is off (lazy=False)
- name: Unit 1
channel: Extracellular
units: uV
amplitude: [50, 150]
firing_rates: # used only if fast loading is off (lazy=False)
- name: Unit 1
kernel: GaussianKernel
sigma: 1.5 # sec
The elephant package’s instantaneous_rate
function is used for calculating
firing rates. See elephant.kernels
for the names of kernel classes that
may be used with the kernel
parameter. neurotic provides an additional
kernel, CausalAlphaKernel
, which may also be used. The
sigma
parameter is passed as an argument to the kernel class and should be
given in seconds.
The rate calculation function and kernel classes are sourced from
neurotic._elephant_tools
, rather than the elephant package itself, to
avoid requiring elephant as a package dependency.
Firing Frequency Burst Detectors¶
If spike trains were generated using
Amplitude Discriminators, imported from
tridesclous Spike Sorting Results, or included in the data_file
, a simple
burst detection algorithm that relies on instantaneous firing rate thresholds
can be run to detect periods of intense activity. Note that burst detectors are
only applied if fast loading is off (lazy=False
).
Detected bursts are plotted as epochs. Colors are inherited from
amplitude_discriminators
, if they are provided there.
Burst detectors are specified in metadata like this:
my favorite dataset:
data_file: data.axgx
# etc
amplitude_discriminators: # used only if fast loading is off (lazy=False)
- name: Unit 1
channel: Extracellular
units: uV
amplitude: [50, 150]
burst_detectors: # used only if fast loading is off (lazy=False)
- spiketrain: Unit 1
name: Unit 1 burst # optional, used for customizing output epoch name
thresholds: [10, 8] # Hz
The algorithm works by scanning through the spike train with a name matching
spiketrain
(in this example, the spike train generated by the “Unit 1”
amplitude discriminator). When the instantaneous firing frequency (IFF; note
this is NOT the same as the smoothed firing rate, but rather the inverse of the inter-spike
interval) exceeds the first threshold given (e.g., 10 Hz), a burst of activity
is determined to start. After this, at the first moment when the IFF drops
below the second threshold (e.g., 8 Hz), the burst is determined to end. After
scanning through the entire spike train, many bursts that meet these criteria
may be identified.
Note that in general the end threshold should not exceed the start threshold; this would essentially be the same as setting the start and end thresholds both to the greater value.
Rectified Area Under the Curve (RAUC)¶
One way to simplify a high-frequency signal is by plotted a time series of the
rectified area under the curve (RAUC). Note that RAUCs are calculated only if
fast loading is off (lazy=False
).
For each signal, the baseline (mean or median) is optionally subtracted off.
The signal is then rectified (absolute value) and divided into non-overlapping
bins of fixed duration. Finally, the integral is calculated within each bin.
The result is a new time series that represents the overall activity of the
original signal. RAUC time series are plotted separately from the original
signals in a second tab. Colors are inherited from plots
, if they are
provided there.
The choice of baseline is controlled by the rauc_baseline
metadata
parameter, which may have the value None
(default), 'mean'
, or
'median'
. The size of the bins determines how smooth the RAUC time series
is and is set by rauc_bin_duration
, given in seconds. If
rauc_bin_duration
is not specified (default None
), RAUC time series
will not be calculated.