SciPy

SciPy is a collection of mathematical algorithms and convenience functions built on NumPy. It provides essential tools for scientific and technical computing, particularly for signal processing, optimization, and statistics.

Why SciPy?

• Comprehensive: Covers most scientific computing needs
• Efficient: Optimized implementations of algorithms
• Well-Tested: Reliable, production-ready code
• Integration: Works seamlessly with NumPy
• Modular: Import only what you need

Key Modules

Signal Processing (scipy.signal)

Essential for neural data analysis:

from scipy import signal
import numpy as np

# Filter a neural signal
b, a = signal.butter(4, [1, 100], 'bandpass', fs=1000)
filtered = signal.filtfilt(b, a, neural_data)

# Compute power spectral density
freqs, psd = signal.welch(lfp_signal, fs=1000)

# Find peaks in data
peaks, _ = signal.find_peaks(trace, height=threshold)

Fourier Transforms (scipy.fft)

from scipy.fft import fft, fftfreq

# Frequency analysis
spectrum = fft(signal)
frequencies = fftfreq(len(signal), 1/sampling_rate)

Optimization (scipy.optimize)

from scipy.optimize import minimize, curve_fit

# Fit function to data
def model(x, a, b):
    return a * np.exp(-b * x)

params, _ = curve_fit(model, x_data, y_data)

Statistics (scipy.stats)

from scipy.stats import ttest_ind, pearsonr

# Statistical tests
t_stat, p_value = ttest_ind(group1, group2)

# Correlation
r, p = pearsonr(neural_activity, behavior)

Interpolation (scipy.interpolate)

from scipy.interpolate import interp1d

# Resample data
f = interp1d(old_times, values, kind='cubic')
new_values = f(new_times)

Common Use Cases in Neuroscience

LFP Analysis

• Bandpass filtering for specific frequency bands
• Hilbert transform for phase and amplitude
• Spectrogram for time-frequency analysis
• Coherence between signals

Spike Analysis

• Convolve with kernels for firing rate estimation
• Detect spike times from continuous recordings
• Cross-correlation for spike-field coupling

Calcium Imaging

• Deconvolve calcium traces to infer spikes
• Baseline correction and normalization
• Detect transient events

Getting Started

Install SciPy:

conda install scipy
# or
pip install scipy

Basic example:

from scipy import signal
import numpy as np
import matplotlib.pyplot as plt

# Generate noisy signal
t = np.linspace(0, 1, 1000)
clean = np.sin(2 * np.pi * 5 * t)
noisy = clean + 0.5 * np.random.randn(len(t))

# Filter signal
sos = signal.butter(4, 10, 'low', fs=1000, output='sos')
filtered = signal.sosfilt(sos, noisy)

# Plot
plt.plot(t, noisy, alpha=0.3, label='Noisy')
plt.plot(t, filtered, label='Filtered')
plt.legend()
plt.show()

Tips

• Use signal.butter with filtfilt for zero-phase filtering
• Check Nyquist frequency before filtering (fs/2)
• Use signal.welch instead of raw FFT for power spectra
• Choose appropriate window functions for spectral analysis
• Use scipy.stats for statistical tests before Pingouin
• Consider scipy.ndimage for image processing tasks