SingleAnalyst

Introduction

SingleAnalyst is an integrated platform for single-cell RNA-seq data analysis, focusing on the cell type assignment problem. SingleAnalyst implemented various quality control, normalization and feature selection methods for data preprocessing, and featured a k-nearest neighbors based cell type annotation and assignment methods. It also extended such method to large scale single cell RNA-seq data by introducing several approximate algorithms.

SingleAnalyst consists of three part: data preprocessing, data inspecting and knn based cell type assignment

Data preprocessing: SingleAnalyst implemented multiple quality control, data normalization and feature selection methods, which are conventionally applied in single-cell RNA-seq analysis.
Data inspecting: SingleAnalyst employed a couple of useful visualization functions for data investigation, as well as a embedded neighbor density based cluster method.
kNN based cell type assignment: SingleAnalyst implemented a k-nearest neighbors based cell type annotation methods. Furthermore, for large-scale single cell RNA-seq data analysis several approximate nearest neighbors methods were deployed, providing the ability to deal with data of variety scale.

Requirement

python3 >= 3.6
linux

Install

Install some dependencies by Anaconda or system's package manager (as pip did not work properly for those packages)
```
conda install numpy bitarray
conda install faiss-cpu -c pytorch
```

Download SingleAnalyst

git clone git@github.com:bm2-lab/Singleanalyst.git

Install SingleAnalyst
```
pip install ./Singleanalyst
```

Usage

Data preprocessing

Read data

Read data, and create a singleCellData object.

from SingleAnalyst.basic import indexedList, infoTable, singleCellData

gene_info = indexedList(gene_list)
cell_info = infoTable(
    ['cell_list', 'cell_type'],
    [cell_list, cell_type_list])
ex_m =  np.loadtxt('expression',delimiter="\t", skiprows=1)
dataset = singleCellData(ex_m, gene_info, cell_info)

Or, read from saved data

import SingleAnalyst

datapath = 'example_data'
data_set = SingleAnalyst.dataIO.read_data_mj(datapath)

After data was loaded, there are several tools for visually inspected the data

SingleAnalyst.vis.plot_g_e(dataset, log=True)
SingleAnalyst.vis.dist_plot(dataset)

Quality control

Filter out low quality data

f1 = SingleAnalyst.filter.minGeneCellfilter()
f2 = SingleAnalyst.filter.minCellGenefilter()

dataset = dataset.apply_proc(f1)
dataset = dataset.apply_proc(f2)

Normalization

Data normalization

norm = SingleAnalyst.normalization.logNormlization()
dataset.apply_proc(norm)

Feature selection

Select informative features.

s1 = SingleAnalyst.selection.dropOutSelecter(num_features=500)
s2 = SingleAnalyst.selection.highlyVarSelecter(num_features=500)
s3 = SingleAnalyst.selection.randomSelecter(num_features=500)

dataset.apply_proc(s1)

The selected features can be visualized

# random pick one feature
one_f = np.random.choice(np.arange(dataset.gene_num))
gn = dataset.index_to_gene([one_f])
v_plot1 = scr.vis.gene_violinplot(dataset, gn[0])

Index build and similarity-based search

For illustration purpose, we split data for test

train_d, test_d = SingleAnalyst.process.tt_split(dataset)
refdata = SingleAnalyst.RefData.queryData(train_d)
q_xdata = SingleAnalyst.RefData.queryData(test_d)

Build index for reference data

nn_indexer = SingleAnalyst.index.faiss_baseline_nn()

index = SingleAnalyst.index.indexRef(refdata, nn=nn_indexer)

kNN search and cell type annotation

qxm = q_xdata.get_qxm(gene_list=index.gene_ref.get_list())

res = index.get_predict(qxm=qxm)

# visually inspect knn result  
i_qx = qxm[19,:]
nnf = index.get_knn_vis(i_qx)

Contacts

yuyifei@tongji.edu.cn or qiliu@tongji.edu.cn