49 GLMs applied to neurons and aggression

| Download notebook

import numpy as np
import scipy.io

import sklearn.linear_model

import cmdstanpy
import arviz as az

import bebi103

import bokeh.io
import bokeh.plotting
bokeh.io.output_notebook()

Loading BokehJS ...

In this lesson, I will demonstrate the use of a GLM in practice. Specifically, we will perform a logistic regression using the data from Remedios, et al. that we have already encountered in ?sec-matlab-files.

Now that we are modeling these data in earnest, it is important to note some facts from the paper about the data set.

Sampling rate of 30 Hz.
The fluorescence reported for each neuron is the relative fluorescent change, \((F(t) - F_0)/F_0\), were \(F_0\) is the average fluorescent intensity over all frames.
Scoring of attack behavior done by human.

We have already loaded and visualized the data set in ?sec-matlab-files. I will expose the code to load in the data set, but hide the visualization code.

data = scipy.io.loadmat('../data/hypothalamus_calcium_imaging_remedios_et_al.mat')
neural_data = data['neural_data'].transpose()  # Row is time index, column is neuron
attack_vector = data['attack_vector'].flatten()
sex_vector = data['sex_vector'].flatten()

# Time points at 30 Hz sampling
t = np.arange(len(attack_vector)) / 30

Code

p_attack = bokeh.plotting.figure(
    frame_height=50,
    frame_width=600,
    x_axis_label=None,
    y_axis_label='attack',
    y_range=['no', 'yes'],
    toolbar_location=None,
)
attack_color = ['tomato' if a else 'gray' for a in attack_vector]
p_attack.scatter(t, ['yes' if a else 'no' for a in attack_vector], size=2, color=attack_color)
p_attack.xaxis.major_label_text_font_size = '0pt'

p_sex = bokeh.plotting.figure(
    frame_height=50,
    frame_width=600,
    x_axis_label=None,
    y_axis_label='sex',
    y_range=['m', 'f'],
    toolbar_location=None,
)
sex_color = ['orchid' if s else 'dodgerblue' for s in sex_vector]
p_sex.scatter(t, ['f' if s else 'm' for s in sex_vector], size=2, color=sex_color)
p_sex.xaxis.major_label_text_font_size = '0pt'

p_neuron = bebi103.image.imshow(
    neural_data.transpose(),
    frame_height=150,
    frame_width=600,
    x_axis_label="time (s)",
    y_axis_label="neuron",
    interpixel_distance=1/30,
)

p_neuron.yaxis.major_label_text_font_size = '0pt'

p_attack.x_range = p_neuron.x_range
p_sex.x_range = p_neuron.x_range

bokeh.io.show(bokeh.layouts.column(p_sex, p_attack, p_neuron))