Clone or download
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
R/FiRE Sync gitFiRE with cranFiRE Jan 5, 2019
data Add files via upload Oct 11, 2018
example Add files via upload Nov 2, 2018
image Add files via upload Oct 11, 2018
python/FiRE Delete fireInstall.log Oct 11, 2018
utils Add files via upload Nov 2, 2018
COPYING Add files via upload Oct 11, 2018
INSTALL Removing custom-boost dependency from R installation. Jan 5, 2019
README.md added instruction of virtual environment. Jan 18, 2019
_config.yml Set theme jekyll-theme-dinky Oct 10, 2018

README.md

FiRE - Finder of Rare Entities

Update: FiRE is now available via cran. install FiRE using

install.packages('FiRE')

Contents

Introduction
External dependencies
Installation
Python Package
-Prerequisites
-Installation Steps
-Usage
R Package
-Prerequisites
-Installation Steps
-Usage
Publication
Copyright
Patent

Introduction

Tested on Ubuntu 14.04 and Ubuntu 16.04.

All results in manuscript have been generated using python module.

FiRE is available for python and R. Required versions and modules for both are as mentioned below. cpp modules are necessary for both of them.

External Dependencies

Following packages are required to run/install the FiRE software.

Required cpp modules

    g++ >= 4.8.4
    boost >= 1.54.0

FiRE only needs <boost/random.hpp> from boost. So, full installation is not necessary. It can be downloaded from boost.org and used as is.

Installation

    [sudo] ./INSTALL [ --boost-path <boost-path> | --log-file <log-file> | --inplace | --py | --R | --help ]
    [sudo] ./UNINSTALL_python
    [sudo] ./UNINSTALL_R

    --boost-path <boost-path>  : python        : Path to boost-library, if boost is not installed at default location, this value needs to be provided.
    --inplace                  : python        : Required only for python, if set, inplace build will be run and resulting lib will be stored in python/FiRE.
    --log-file <log-file>      : python        : Required only for python, ignored with --inplace set.
    --py                       : python        : Install FiRE in python environment.
    --R                        : R             : Install FiRE in R environment.
    --help                     : python | R    : Display this help.

    Info:

    UNINSTALL_[python | R] files are generated upon installation.

Typically, FiRE module takes a few seconds to install. A snippet of installation time taken by FiRE (in seconds) on a machine with Intel® Core™ i5-7200U (CPU @ 2.50GHz × 4), with 8GB memory, and OS Ubuntu 16.04 LTS is as follows

real 2.92
user 2.73
sys 0.18

Python Package

Prerequisites

Required python modules

    python 2.7
    # [EDIT] python 3 can also be used with standard installation. 
    # however, with --inplace option uninstallation may fail. 
    # As a workround generated .so file can be removed manually if uninstallation of FiRE is desired.

For FiRE module

    cython >= 0.23.4
    distutils >= 2.7.12

For preprocessing

    numpy >= 1.13.3
    pandas >= 0.20.3
    statsmodels >= 0.8.0

For demo

    gzip >= 1.2.11 (zlib)
    scipy >= 1.1.0
    matplotlib >= 2.1.0
    cmocean >= 1.2
    sklearn >= 0.19.1

Installation Steps

with virtual environment avoid using sudo. (Thanks to chenxofhit)

[sudo] chomd +x ./INSTALL

If boost is installed at default location

[sudo] ./INSTALL --py

If boost is installed at custom location

[sudo] ./INSTALL --boost-path <full-path> --py

Example:

[sudo] ./INSTALL --boost-path $HOME/boost/boost_1_54_0 --py

Above installation steps will generate fireInstall.log file. It is advisable to keep this file, since it will be needed for uninstallation. Name of the log file can be modified during installation.

./INSTALL --log-file <log-file-name> --py

Above steps will install FiRE at the default location.

For inplace installation

./INSTALL --inplace --py

Uninstallation of FiRE Software.

[sudo] ./UNINSTALL_python

Usage

Run demo from FiRE directory as follows

python example/jurkat_simulation.py

Since data (data/jurkat_two_species_1580.txt.gz) is large, this may require large amount of RAM to load and pre-process. We have also providee pre-processed data (data/preprocessedData_jurkat_two_species_1580.txt.gz). Pre-processing was done using the script present in utils/preprocess.py. Demo using this data as follows

python example/jurkat_simulation_small.py

Small demo takes seconds to complete. Exact time taken by the demo on a machine with Intel® Core™ i5-7200U (CPU @ 2.50GHz × 4), with 8GB memory, and OS Ubuntu 16.04 LTS is as follows

Loading preprocessed Data : 1.850723s
Running FiRE : 1.134673s

Total Demo time:

real 4.33
user 3.55
sys 0.76

Step-by-step description of full demo (example/jurkat_simulation.py) is as follows

  1. Load libraries

import sys
sys.path.append('utils')

import numpy as np
import gzip
from scipy import stats

import preprocess as pp
import misc
import FiRE
  1. Load Data in current environment.

#Data matrix should only consist of values where rows represent cells and columns represent genes.

with gzip.GzipFile('data/jurkat_two_species_1580.txt.gz', 'r') as fid:
    data = np.genfromtxt(fid)

data = data.T #Samples * Features

labels = np.genfromtxt('data/labels_jurkat_two_species_1580.txt', dtype=np.int) #Cells with label '1' represent abundant, while cells with label '2' represent rare.

Data Pre-processing

  1. Call function ranger_preprocess for selecting thousand variable genes.

#Genes
genes = np.arange(1, data.shape[1]+1) #It can be replaced with original gene names

#Filter top 1k genes
preprocessedData, selGenes = pp.ranger_preprocess(data, genes, optionToSave=True, dataSave=outputFolder)
Parameter Description Required or Optional Datatype Default Value
data Data for processing Required np.array [nCells, nGenes] -
genes Names of Genes Required np.array [nGenes] -
ngenes_keep Number of genes to keep Optional integer 1000
dataSave Path to save results Optional string Current working Directory (Used only when optionToSave is True)
optionToSave Save processed output or not Optional boolean False(Does not save)
minLibSize Minimum number of expressed features Optional integer 0
verbose Display progress Optional boolean True(Prints intermediate results)
'''
Returned Value :
    preprocessedData : processed data matrix (log2 transformed) : np.array [nCells, nVariableGenes]
    selGenes         : Names of thousand variable genes selected : np.array [nVariableGenes]
'''
  1. Create model of FiRE.

model = FiRE.FiRE(L=100, M=50, H=1017881, seed=5489, verbose=0)
Parameter Description Required or Optional Datatype Default Value
L Total number of estimators Required int -
M Number of features to be randomly sampled for each estimator Required int -
H Number of bins in hash table Optional int 1017881
seed Seed for random number generator Optional unsigned int 5489
verbose Controls verbosity of program at run time (0/1) Optional int 0 (silent)
  1. Apply model to the above dataset.

model.fit(preprocessedData)
  1. Calculate FiRE score of every cell.

score = np.array(model.score(preprocessedData))
'''
Returned Value :
    score : FiRE score of every cell : np.array[nCells]

Higher values of FiRE score represent rare cells.
'''
  1. Select cells with higher values of FiRE score, that satisfy IQR-based thresholding criteria.

q3 = np.percentile(score, 75)
iqr = stats.iqr(score)
th = q3 + 1.5*iqr

indIqr = np.where(score >= th)[0]

dataSel = preprocessedData[indIqr,:] #Select subset of rare cells

#Create a file with binary predictions
predictions = np.zeros(data.shape[0])
predictions[indIqr] = 1 #Replace predictions for rare cells with '1'.
  1. Access to model parameters.

Sampled dimensions can be accessed via

# type : 2d list
# shape : L x M
model.dims

Chosen thresholds can be accessed via

# type : 2d list
# shape : L x M
model.thresholds

Weights can be accessed via

# type : 2d list
# shape : L X M
model.weights

Hash tables can be accessed via

# type : 3d list
# shape : L x H x <dynamic>
# <dynamic> : as per number of samples in a bin (H) for a given estimator (L).
model.bins
  1. FiRE recovers artifitially planted rare cells (Figure).

(a) t-SNE based 2D embedding of the cells with color-coded identities (b) FiRE score intensities plotted on the t-SNE based 2D map. (c) Rare cells detected by FiRE.

R Package

Prerequisites

Required R modules

    R >= 3.2.0

For FiRE module

    Rcpp >= 0.12.19
    BH >= 1.66

For preprocessing and demo

    Matrix >= 1.2.14
    plyr >= 1.8.4

Installation Steps

[sudo] chomd +x ./INSTALL

Installation of FiRE Software.

[sudo] ./INSTALL --R

Uninstallation of FiRE Software.

[sudo] ./UNINSTALL_R

Usage

Run demo from FiRE directory as follows

Rscript example/jurkat_simulation.R

Since data (data/jurkat_two_species_1580.txt.gz) is large, this may require large amount of RAM to load and pre-process. We have also providee pre-processed data (data/preprocessedData_jurkat_two_species_1580.txt.gz). Pre-processing was done using the script present in utils/preprocess.R. Demo using this data as follows

Rscript example/jurkat_simulation_small.R

Small demo takes seconds to complete. Exact time taken by the demo on a machine with Intel® Core™ i5-7200U (CPU @ 2.50GHz × 4), with 8GB memory, and OS Ubuntu 16.04 LTS is as follows

Total Demo time:

real 4.11
user 3.16
sys 1.13

Step-by-step description of full demo (example/jurkat_simulation.R) is as follows

  1. Load libraries

library('FiRE')
source('utils/preprocess.R')
  1. Load Data in current environment.

#Read data
data <- read.table(gzfile('data/jurkat_two_species_1580.txt.gz'))
data <- t(data) #Samples * Features

#Read Labels
labels <- read.table('data/labels_jurkat_two_species_1580.txt') #Cells with label '1' represent abundant, while cells with label '2' represent rare.

#Genes
genes <- c(1:dim(data)[2]) #It can be replaced with original gene names

data_mat <- list(mat=data, gene_symbols=genes)
  1. Call function ranger_preprocess for selecting thousand variable genes.

preprocessedList <- ranger_preprocess(data_mat)
preprocessedData <- as.matrix(preprocessedList$preprocessedData)
Parameter Description Required or Optional Datatype Default Value
data_mat List consisting of data for processing and gene symbols Required list(mat=data, gene_symbols=genes) -
ngenes_keep Number of genes to keep Optional integer 1000
dataSave Path to save results Optional string Current working Directory (Used only when optionToSave is True)
optionToSave Save processed output or not Optional boolean False(Does not save)
minLibSize Minimum number of expressed features Optional integer 0
verbose Display progress Optional boolean True(Prints intermediate results)
  1. Create model of FiRE.

# model <- new(FiRE::FiRE, L, M, H, seed, verbose)
model <- new(FiRE::FiRE, 100, 50, 1017881, 5489, 0)
Parameter Description Required or Optional Datatype Default Value
L Total number of estimators Required int -
M Number of features to be randomly sampled for each estimator Required int -
H Number of bins in hash table Optional int 1017881
seed Seed for random number generator Optional int 5489
verbose Controls verbosity of program at run time (0/1) Optional int 0 (silent)
  1. Apply model to the above dataset.

model$fit(preprocessedData)

Acceptable datatype is of matrix class and of type double (Numeric matrix).

  1. Calculate FiRE score of every cell.

# Returns a numeric vector
score <- model$score(preprocessedData)
  1. Select cells with higher values of FiRE score, that satisfy IQR-based thresholding criteria.

#Apply IQR-based criteria to identify rare cells for further downstream analysis.
q3 <- quantile(score, 0.75)
iqr <- IQR(score)
th <- q3 + (1.5*iqr)

#Select indexes that satisfy IQR-based thresholding criteria.
indIqr <- which(score >= th)

#Create a file with binary predictions
predictions <- integer(dim(data)[1])
predictions[indIqr] <- 1 #Replace predictions for rare cells with '1'.
  1. Access to model parameters.

Sampled dimensions can be accessed via

# type : Integer matrix
# shape : L x M
model$d

Chosen thresholds can be accessed via

# type : Numeric matrix
# shape : L x M
model$ths

Weights can be accessed via

# type : Numeric matrix
# shape : 0 x 0
model$w

Hash tables can be accessed via

# type : List
# shape : L x H x <dynamic>
# <dynamic> : as per number of samples in a bin (H) for a given estimator (L).
model$b

Publication

Jindal, A., Gupta, P., Jayadeva and Sengupta, D., 2018. Discovery of rare cells from voluminous single cell expression data. Nature communications, 9(1), p.4719. DOI: https://doi.org/10.1038/s41467-018-07234-6

Copyright

This software package is distributed under GNU GPL v3.

Patent