scBKAP: a single-cell RNA-Seq data clustering pipeline

Overview

scBKAP, the cornerstone of which is a single-cell bisecting K-means clustering method based on an autoencoder network and a dimensionality reduction model MPDR. Specially, scBKAP utilizes an autoencoder network to reconstruct gene expression values from scRNA-Seq data to alleviate the dropout issue, and the MPDR model composed of the M3Drop feature selection algorithm and the PHATE dimensionality reduction algorithm to reduce the dimensions of reconstructed data. The dimension-ality-reduced data are then fed into the bisecting K-means clustering algorithm to identify the clusters of cells.

Requirement:

python = 3.6
numpy = 1.16.3
scipy = 1.4.1
pandas = 0.25.3
scikit-learn = 0.22.1
tensorflow = 1.13.1
matplotlib = 3.0.3
graphtools = 1.5.2
R = 3.6

Dataset:

We only provide one single-cell RNA-seq dataset Ting, other datasets can be obtained from https://hemberg-lab.github.io/scRNA.seq.datasets/

Usage:

Input

The input of scBKAP should be a csv file (row: genes, col: cells).

Run

Run the genefilter.py to filter the data， the data is the scRNA-seq data matrix:

from main import *

filte('ting', 'ting_filter')

Reconstruct the data:

X_con = autorunner('ting_filter', 1000, 800, 200, 'ting_auto')

# autorunner(data_name, epochs, h1, h2, args_name)
# epochs is Number of iterations， default value 1000
# h1 and h2 are the hidden layers， default value 800, 200

Use the M3drop:

library(M3drop)
b <- read.csv('ting_auto.csv',header = F)
Normalized_data <- M3DropCleanData(b, 
                                   is.counts=TRUE, 
                                   min_detected_genes=112)
#the min_detected_genes will changes with the different dataset and the different reconstructed data, because of different expression values
c <- Normalized_data$data
c <- t(c)
write.csv(c,'data_m3.csv',row.names = F, col.names = F)

clust the data:

y_pred = clust('ting_m3', 'ting_label', 20, 5)

# clust(data_path, label_path, pca_com, phate_com)
# If the number of cells is more than 800, the `pca_com` is 100, otherwise is 20
# Default value of phate_com is 5

Output

ting_filter.csv The filtered data.
ting_auto.csv The reconstructed data.
ting_m3.csv The data selected by M3drop.
y_pred The predicted label by scBKAP.

xuyaokui/scBKAP