Welcome to Harissa’s 🌶 documentation!#
Contents:
This is a Python package for both simulation and inference of gene regulatory networks from single-cell data. Its name comes from ‘HARtree approximation for Inference along with a Stochastic Simulation Algorithm.’ It was implemented in the context of a mechanistic approach to gene regulatory network inference from single-cell data, based upon an underlying stochastic dynamical model driven by the transcriptional bursting phenomenon.
Main functionalities:
Network inference interpreted as calibration of a dynamical model;
Data simulation (typically scRNA-seq) from the same dynamical model.
Other available tools:
Basic GRN visualization (directed graphs with positive or negative edge weights);
Binarization of scRNA-seq data (using gene-specific thresholds derived from the calibrated dynamical model).
The current version of Harissa has benefited from improvements introduced within Cardamom, which can be seen as an alternative method for the inference part. The two inference methods remain complementary at this stage and may be merged into the same package in the future. They were both evaluated in a recent benchmark.
Installation#
Harissa can be installed using pip:
pip install harissa
This command will also check for all required dependencies (see below)
and install them if necessary.
If the installation is successful, all scripts in the
tests
folder should run smoothly
(note that network4.py must be run before test_binarize.py).
Basic usage#
from harissa import NetworkModel
model = NetworkModel()
# Inference
model.fit(data)
# Simulation
sim = model.simulate(time)
Here data should be a two-dimensional array of single-cell gene expression counts,
where each row represents a cell and each column represents a gene,
except for the first column, which contains experimental time points.
A toy example is:
import numpy as np
data = np.array([
#t g1 g2 g3
[0, 4, 1, 0], # Cell 1
[0, 5, 0, 1], # Cell 2
[1, 1, 2, 4], # Cell 3
[1, 2, 0, 8], # Cell 4
[1, 0, 0, 3], # Cell 5
])
The time argument for simulations is either a single time or a list of time points.
For example, a single-cell trajectory (not available from scRNA-seq)
from t = 0h to t = 10h can be simulated using:
time = np.linspace(0, 10, 1000)
The sim output stores mRNA and protein levels as attributes
sim.m and sim.p,
respectively (each row is a time point and each column is a gene).
About the data#
The inference algorithm specifically exploits time-course data, where single-cell profiling is performed at a number of time points after a stimulus (see this paper for an example with real data). Each group of cells collected at the same experimental time t k forms a snapshot of the biological heterogeneity at time t k. Due to the destructive nature of the measurement process, successive snapshots are made of different cells. Such data is therefore different from so-called ‘pseudotime’ trajectories,