Matrix processing

usage: bin/matrix-process.R [-h] -s STEP -i INPUT -c CLASS -b BATCH
                            --filterout FILTEROUT --imputeout IMPUTEOUT
                            --normalizeout NORMALIZEOUT --batchremoveout
                            BATCHREMOVEOUT [--filtercount NUMBER]
                            [--filtersample NUMBER] [--imputemethod STRING]
                            [--imputecluster NUMBER] [--imputenum NUMBER]
                            [--imputecutoff NUMBER] [--imputealpha NUMBER]
                            [--normmethod STRING] [--normtopk NUMBER]
                            [--cvthreshold NUMBER] [--removetype STRING]
                            [--refergenefile STRING] [--batchmethod STRING]
                            [--batchindex INT] [-p NUMBER]

optional arguments:
  -h, --help            show this help message and exit
  -s STEP, --step STEP  which step to run
  -i INPUT, --input INPUT
                        input expression matrix file
  -c CLASS, --class CLASS
                        input class info file
  -b BATCH, --batch BATCH
                        input batch info file
  --filterout FILTEROUT
                        output filter path
  --imputeout IMPUTEOUT
                        output imputation path
  --normalizeout NORMALIZEOUT
                        output normalization file
  --batchremoveout BATCHREMOVEOUT
                        output batchremoved file
  --filtercount NUMBER  filter by counts of a gene [default = 5]
  --filtersample NUMBER
                        filter by counts of sample above certain counts of a
                        gene [default = 10]
  --imputemethod STRING
                        the imputation algorithm to use [default =
                        scimpute_count]
  --imputecluster NUMBER
                        cluster number in scImpute [default = 5]
  --imputenum NUMBER    number in viper [default = 5000]
  --imputecutoff NUMBER
                        cutoff in viper [default = 0.5]
  --imputealpha NUMBER  alpha in viper [default = 0.1]
  --normmethod STRING   the normalization algorithm to use [default = SCNorm]
  --normtopk NUMBER     top K feature as scale factor [default = 20]
  --cvthreshold NUMBER  coefficient variance threshold of reference gene,
                        filter ref gene with CV bigger than [default = 0.5]
  --removetype STRING   remove some time of RNA for normalization scale factor
                        calculation [default = miRNA,piRNA]
  --refergenefile STRING
                        reference gene file path [default = None]
  --batchmethod STRING  the batch removal algorithm to use [default = RUV]
  --batchindex INT      batch index to select which batch to use [default = 1]
  -p NUMBER, --processors NUMBER
                        Number of processors to use. This option is useful on
                        multicore *nix or Mac machine only, when performing
                        multiple runs (nrun > 1) [default = 1]

  • Some parameters:

  • Example:

Plots of results

We provide a jupyter notebook to plot the results of matrix processing. notebook

environment

color

RLE

heterogeneity

uca_knn

understand UCA

understand mkNN

first alignment score

Alignment Score=1k−kN(k−x‾)\text{Alignment\ Score} = \frac{1}{k-\frac{k}{N}}(k-\overline{x})

其中$k$是最近邻算法（k nearest-neighbors, kNN）的前$k$个最近邻，$\overline{x}$是样本周围的样本同属一个批次的数量的平均，$N$表示样本数。 当两个批次样本完全分开时，$k=\overline{x}$，$\text{Alignment Score}=0$；当两个批次样本完全混杂时，比例因子$\frac{1}{k-\frac{k}{N}}$作用下，$\text{Alignment Score}$接近1。 exSEEK提出了适用于多种批次的mkNN指标，该指标由由史斌斌首次提出。

mkNN

Alignment Score=1−x‾−kNk−kN\text{Alignment\ Score} = 1-\frac{\overline{x}-\frac{k}{N}}{k-\frac{k}{N}}
mkNN=1−1B∑b=1Bx‾b−kNb/(N−1)min⁡(k,Nb)−kNb/(N−1)\text{mkNN}=1-\frac{1}{B} \sum_{b=1}^{B} \frac{\overline{x}_{b}-k N_{b} /(N-1)}{\min \left(k, N_{b}\right)-k N_{b} /(N-1)}

其中，$b$表示批次，$B$为批次数量，$N_b$是批次$b$下样本的数量。批次效应越明显，该指标越接近0。

PCA

variance explained

PreviousConfiguration File ReferenceNextRun on a cluster

Last updated

Was this helpful?