Log2ratio transformation, feature selection and simple random sampling on a gene expression matrix

l2r_fsel_srs.py

import pandas as pd
import numpy as np
import sys
import random as rnd

csv = sys.argv[1]
out = sys.argv[2]

df = pd.read_table(csv, sep='\t', index_col=0)

# log2
df = np.log2(df)

# remove genes with more than 5% of samples below the 5th percentile
thres = np.percentile(df, 5)
quantile = df.quantile(0.05, axis=1)
genes = quantile[quantile > thres].index.tolist()
df = df.ix[genes, :]

# ratio
median = df.median(axis='columns')
tcga = df.sub(median, axis='index')

# feature selection, most variant genes
Kgenes = 1000

std = df.std(axis='columns')
rank = std.rank(ascending=False)
genes = rank[rank < Kgenes].index.tolist()

df = df.ix[genes, :]

# random sampling 100 samples
Ksamples = 100
samplesix = rnd.sample(range(len(newcolumns)), Ksamples)
samples = [newcolumns[i] for i in samplesix]

df = df.ix[:, samples]

# output
df.to_csv(out, sep='\t')