RLFimpute

Imputation for scRNA-seq data based on the reinforcement learning framwork (RLF).

Brief Introduction

• RLFimpute is a novel imputation algorithm based on the reinforcement learning framework (RLF) to deal with dropouts in scRNA-seq data. We see dropout as agent, current matrix as environment, the process of imputation as state changes, and Davies-Bouldin Index as reward.
  
• The input of RLFimpute is read count matrix data (genes x cells). It preprocessed the data and applies RLFimpute to get the candidate sets of dropouts and then get the imputed data (genes x cells).
  
• We have tested RLFimpute on Matlab2018a software on different operating systems(Windows, Linux, Mac OS). RLFimpute enables GPU acceleration inaddition to CPU computing as long as the GPUs are available.
  

Dependences

RLFimpute depends on matlab's parfor parallel computing, make sure your matlab version can use parfor before running RLFimpute.
Before using parfor, you must first configure and start the parallel computing pool.
Parpool function provides parallel pool configuration and opening functions.

p=parpool

If your parfor works, you will get the information about NumWorkers, IdleTimeout, SpmdEnabled, etc.

Implementation

Running the script would read the rawdata matrix, processes it, call RLFimpute and get the recovered matrix.

parfor_main.m

Data visualization

PCA visualization:

visulaization.m

T-sne visualization:

addpath(genpath('TSNE'));
mappedX = tsne(rawcount_imp',[],2,50,30);
gscatter(mappedX(:,1),mappedX(:,2),X);

Calculation of clustering evaluation indices

Calculate rand index:

ri = rand_index(p, p1);

Calculate adjusted rand index:

ari = rand_index(p, p1, 'adjusted');

GPU acceleration

When the data is too large, you can run RLFimpute on GPUs for acceleration.
Get the information about your GPU:

gpuDevice

Move data from CPU to GPU using function: gpuArray, usage:

rawdata = gpuArray(data);

After calculation, the imputed data is moved out to the CPU storage using function: gather, usage:

rawdata_imp = gather(rawdata_norm);