NMF（非负矩阵分解）分子分型

Non-Negative Matrix Factorization (NMF).

Find two non-negative matrices, i.e. matrices with all non-negative elements, (W, H) whose product approximates the non-negative matrix X. This factorization can be used for example for dimensionality reduction, source separation or topic extraction.

安装NMF

ubuntu

ubuntu上，编译需要

sudo apt install libopenmpi-dev

R

using(pak)

pak::pkg_install("NMF",dependencies=T)

使用

run_nmf(

    exp=exp,

    genelist=c("PCNA","HNRNPK","TRIM28","NPM1","PARK7","HDAC1")

)

exp：表达矩阵，标准化过，但是不能有负数，行是基因，列是样本

#        TCGA-3L-AA1B-01A TCGA-4N-A93T-01A TCGA-4T-AA8H-01A

# MT-CO2          14.77639        15.77524        16.05650

# MT-CO3          15.13540        16.16666        15.84924

# MT-ND4          14.66976        14.80350        15.21889

# MT-CO1          13.98580        14.53619        15.30272

# MT-ATP6        13.53251        14.28397        14.60036

genelist ：基因名向量，基因名需要在exp的行名中，如果为空则使用全部基因

method: 最常用的三种brunet、lee、snmf/r

n_run：运行次数

结果

结果怎么看https://mubu.com/doc/C4gVcgp-G0

图片

图片

R function

run_nmf <- function(

    exp,

    genelist=NULL,

    od = '.',

    n_cluster = 3,

    n_run=30,

    method="brunet",

    cluster_range=2:10,

    seed = 1314,

    cluster_character = "Cluster",

    )

{

    if (!dir.exists(od)) {

        dir.create(od)

    }

    if(!is.null(genelist)){

        exp <- exp[which(rownames(exp) %in% genelist), ]

    }

   

    using(NMF,data.table,tidyverse)

    if(is.numeric(cluster_range)){

        result <- NMF::nmf(exp,

            cluster_range,

            method = method,

            nrun = n_run,

            seed = seed

        )

        plot(result)

        ps(paste0(od, "/ranks.pdf"),w=10,h=10)

    }

    result2 <- NMF::nmf(exp, method = method, rank = n_cluster, seed = seed,nrun = n_run)

    key_gene <- NMF::extractFeatures(result2, 0.5) # 提取关键基因

    fwrite(data.table(key_gene=key_gene),paste0(od,'/key_gene.csv'))

    # 提出亚型

    Cluster <- as_tibble(predict(result2), rownames = "Sample") %>%

        dplyr::rename(Cluster = value) %>%

        dplyr::mutate(Cluster = paste0("Cluster", Cluster))

    fwrite(Cluster, paste0(od, "/NMF_Cluster.csv"))

    consensusmap(result2,

        labRow = NA,

        labCol = NA,

        annCol = data.frame("cluster" = predict(result2)[colnames(exp)])

    )

    ps(paste0(od, "/Cluster.pdf"),w=6,h=6)

    return()

}

ps <- function(filename, plot = FALSE, w = 12, h = 6) {

    if (is.object(plot)) {

        print(plot)

    }

    plot <- recordPlot()

    pdf(file = filename, onefile = T, width = w, height = h)

    replayPlot(plot)

    dev.off()

}

Reference

https://cloud.tencent.com/developer/article/1806266

https://mubu.com/doc/C4gVcgp-G0

https://scikit-learn.org/stable/modules/generated/sklearn.decomposition.NMF.html

https://www.geeksforgeeks.org/non-negative-matrix-factorization/