renge 0.0.2.dev3

Infer gene regulatory networks from time-series single-cell CRISPR data.

RENGE

RENGE infers gene regulatory networks (GRNs) from time-series single-cell CRISPR screening data.

Install

pip install renge

Usage

Network inference

reg = Renge()
A = reg.estimate_hyperparams_and_fit(X, E)
A_qval = reg.calc_qval()

input
E : C x G pandas DataFrame of expression data. The expression data should be normalized and log-transformed.
X : C x (G+1) pandas DataFrame. The rows of X[:, :G] are one-hot vectors indicating the knocked out gene in each cell. The last column of X indicates the sampling time of each cell.

Here,
C : The number of cells.
G : The number of genes included in the GRN.

output
A : G x G pandas DataFrame. (i, j) element of A is a regulatory coefficient from gene j to gene i.
A_qval : G x G pandas DataFrame. (i, j) element of A_qval is a q-value for (i, j) element of A.

Prediction of expression changes after gene knockout

reg = Renge()
reg.estimate_hyperparams_and_fit(X_train, E_train)   # train the model
E_pred = reg.predict(X_pred)

input
E_train : C x G pandas DataFrame of expression data. The expression data should be normalized and log-transformed.
X_train : C x (G+1) pandas DataFrame. The rows of X_train[:, :G] are one-hot vectors indicating the knocked out gene in each cell. The last column of X_train indicates the sampling time of each cell.
X_pred : T x (G+1) pandas DataFrame. The rows of X[:, :G] are real-valued vectors indicating expression change of target gene of perturbation. For knockout or knockdown, values should be negative. The last column of X_pred indicates the time at which expressions are predicted

Here,
T : The number of timepoints where expressions are predicted.

output
E_pred : T x G pandas DataFrame of predicted expression.

Reference

TBD