Accuracy as a function of isotope incorporation & community similarity among replicate microcosms
Note: just adding more levels to the original microBetaDiv simulation
In [1]:
import os
# paths
workDir = '/home/nick/notebook/SIPSim/dev/bac_genome1147/'
buildDir = os.path.join(workDir, 'microBetaDiv_morePerm')
R_dir = '/home/nick/notebook/SIPSim/lib/R/'
fragFile = '/home/nick/notebook/SIPSim/dev/bac_genome1147/validation/ampFrags_kde.pkl'
genome_index = '/var/seq_data/ncbi_db/genome/Jan2016/bac_complete_spec-rep1_rn/genome_index.txt'
In [2]:
import glob
import itertools
import nestly
In [3]:
%load_ext rpy2.ipython
%load_ext pushnote
In [4]:
%%R
library(ggplot2)
library(dplyr)
library(tidyr)
library(gridExtra)
In [5]:
if not os.path.isdir(buildDir):
os.makedirs(buildDir)
%cd $buildDir
In [6]:
## min G+C cutoff
min_GC = 13.5
## max G+C cutoff
max_GC = 80
## max G+C shift
max_13C_shift_in_BD = 0.036
min_BD = min_GC/100.0 * 0.098 + 1.66
max_BD = max_GC/100.0 * 0.098 + 1.66
max_BD = max_BD + max_13C_shift_in_BD
print 'Min BD: {}'.format(min_BD)
print 'Max BD: {}'.format(max_BD)
In [7]:
# making an experimental design file for qSIP
x = range(1,7)
y = ['control', 'treatment']
expDesignFile = os.path.join(buildDir, 'qSIP_exp_design.txt')
with open(expDesignFile, 'wb') as outFH:
for i,z in itertools.izip(x,itertools.cycle(y)):
line = '\t'.join([str(i),z])
outFH.write(line + '\n')
!head $expDesignFile
In [10]:
# building tree structure
nest = nestly.Nest()
# varying params: test
#nest.add('shared_perc', [90, 95])
#nest.add('perm_perc', [5, 10])
#nest.add('rep', range(1,4))
# varying params
nest.add('shared_perc', [80, 85, 90, 95, 100])
nest.add('perm_perc', [30, 40, 50])
nest.add('rep', range(1,11))
## set params
nest.add('percIncorp', [100], create_dir=False)
nest.add('percTaxa', [10], create_dir=False)
nest.add('abs', ['1e9'], create_dir=False)
nest.add('np', [8], create_dir=False)
nest.add('Monte_rep', [100000], create_dir=False)
nest.add('subsample_dist', ['lognormal'], create_dir=False)
nest.add('subsample_mean', [9.432], create_dir=False)
nest.add('subsample_scale', [0.5], create_dir=False)
nest.add('subsample_min', [10000], create_dir=False)
nest.add('subsample_max', [30000], create_dir=False)
nest.add('min_BD', [min_BD], create_dir=False)
nest.add('max_BD', [max_BD], create_dir=False)
nest.add('DBL_scaling', [0.5], create_dir=False)
nest.add('bandwidth', [0.8], create_dir=False)
nest.add('heavy_BD_min', [1.71], create_dir=False)
nest.add('heavy_BD_max', [1.75], create_dir=False)
nest.add('topTaxaToPlot', [100], create_dir=False)
nest.add('padj', [0.1], create_dir=False)
nest.add('log2', [0.25], create_dir=False)
### input/output files
nest.add('buildDir', [buildDir], create_dir=False)
nest.add('R_dir', [R_dir], create_dir=False)
nest.add('genome_index', [genome_index], create_dir=False)
nest.add('fragFile', [fragFile], create_dir=False)
nest.add('exp_design', [expDesignFile], create_dir=False)
# building directory tree
nest.build(buildDir)
# bash file to run
bashFile = os.path.join(buildDir, 'SIPSimRun.sh')
In [11]:
bashFileTmp = os.path.splitext(bashFile)[0] + '_expDesign.sh'
bashFileTmp
Out[11]:
In [12]:
%%writefile $bashFileTmp
#!/bin/bash
echo '#-- Experimental design --#'
echo '# Making an isotope incorporation config file'
echo '## 3 replicate gradients for control & treatment'
SIPSim incorpConfigExample \
--percIncorpUnif {percIncorp} \
--n_reps 3 \
> incorp.config
echo '# Selecting incorporator taxa'
echo '## This is to make the gradient replicates consistent (qSIP finds mean among replicates)'
SIPSim KDE_selectTaxa \
-p {percTaxa} \
{fragFile} \
> incorporators.txt
echo '# Creating a community file (3 replicate control, 3 replicate treatment)'
SIPSim communities \
--richness 950 \
--config incorp.config \
--shared_perc {shared_perc} \
--perm_perc {perm_perc} \
{genome_index} \
> comm.txt
echo '# simulating gradient fractions'
SIPSim gradient_fractions \
--BD_min {min_BD} \
--BD_max {max_BD} \
comm.txt \
> fracs.txt
In [13]:
!chmod 777 $bashFileTmp
!cd $workDir; \
nestrun --template-file $bashFileTmp -d $buildDir --log-file exp_design.log -j 10
In [14]:
%pushnote exp_design complete: $buildDir
In [15]:
bashFileTmp = os.path.splitext(bashFile)[0] + '_SIPSim-pipeline.sh'
bashFileTmp
Out[15]:
In [ ]:
%%writefile $bashFileTmp
#!/bin/bash
echo '#-- SIPSim pipeline --#'
echo '# Adding diffusion'
SIPSim diffusion \
-n {Monte_rep} \
--bw {bandwidth} \
--np {np} \
{fragFile} \
> ampFrags_KDE_dif.pkl
echo '# Adding DBL contamination; abundance-weighted smearing'
SIPSim DBL \
-n {Monte_rep} \
--comm comm.txt \
--commx {DBL_scaling} \
--np {np} \
ampFrags_KDE_dif.pkl \
> ampFrags_KDE_dif_DBL.pkl
echo '# Adding isotope incorporation to BD distribution'
SIPSim isotope_incorp \
-n {Monte_rep} \
--comm comm.txt \
--taxa incorporators.txt \
--np {np} \
ampFrags_KDE_dif_DBL.pkl \
incorp.config \
> ampFrags_KDE_dif_DBL_inc.pkl
echo '# Simulating an OTU table'
SIPSim OTU_table \
--abs {abs} \
--np {np} \
ampFrags_KDE_dif_DBL_inc.pkl \
comm.txt \
fracs.txt \
> OTU_abs{abs}.txt
echo '# Simulating PCR'
SIPSim OTU_PCR \
OTU_abs{abs}.txt \
> OTU_abs{abs}_PCR.txt
echo '# Subsampling from the OTU table (simulating sequencing of the DNA pool)'
SIPSim OTU_subsample \
--dist {subsample_dist} \
--dist_params mean:{subsample_mean},sigma:{subsample_scale} \
--min_size {subsample_min} \
--max_size {subsample_max} \
OTU_abs{abs}_PCR.txt \
> OTU_abs{abs}_PCR_sub.txt
echo '# Making a wide-formatted table'
SIPSim OTU_wideLong -w \
OTU_abs{abs}_PCR_sub.txt \
> OTU_abs{abs}_PCR_sub_w.txt
echo '# Making metadata (phyloseq: sample_data)'
SIPSim OTU_sampleData \
OTU_abs{abs}_PCR_sub.txt \
> OTU_abs{abs}_PCR_sub_meta.txt
#-- removing large intermediate files --#
rm -f ampFrags_KDE_dif.pkl
rm -f ampFrags_KDE_dif_DBL.pkl
rm -f ampFrags_KDE_dif_DBL_inc.pkl
In [ ]:
!chmod 777 $bashFileTmp
!cd $workDir; \
nestrun --template-file $bashFileTmp -d $buildDir --log-file SIPSim_pipeline.log -j 2
In [ ]:
%pushnote SIPSim pipeline complete: $buildDir
In [ ]:
bashFileTmp = os.path.splitext(bashFile)[0] + '_SIPSim-summary.sh'
bashFileTmp
In [ ]:
%%writefile $bashFileTmp
#!/bin/bash
# plotting 'raw' taxon abundances
SIPSimR OTU_taxonAbund \
OTU_abs{abs}.txt \
-r {topTaxaToPlot} \
-o OTU_abs{abs}
# plotting 'sequenced' taxon abundances
SIPSimR OTU_taxonAbund \
OTU_abs{abs}_PCR_sub.txt \
-r {topTaxaToPlot} \
-o OTU_abs{abs}_PCR_sub
In [ ]:
!chmod 777 $bashFileTmp
!cd $workDir; \
nestrun --template-file $bashFileTmp -d $buildDir --log-file SIPSim_summary.log -j 10
In [ ]:
bashFileTmp = os.path.splitext(bashFile)[0] + '_HRSIP.sh'
bashFileTmp
In [ ]:
%%writefile $bashFileTmp
#!/bin/bash
# phyloseq
## making phyloseq object from OTU table
SIPSimR phyloseq_make \
OTU_abs{abs}_PCR_sub_w.txt \
-s OTU_abs{abs}_PCR_sub_meta.txt \
> OTU_abs{abs}_PCR_sub.physeq
## filtering phyloseq object to just 'heavy' fractions
SIPSimR phyloseq_edit \
OTU_abs{abs}_PCR_sub.physeq \
--BD_min {heavy_BD_min} \
--BD_max {heavy_BD_max} \
> OTU_abs{abs}_PCR_sub_filt.physeq
## making ordination
SIPSimR phyloseq_ordination \
OTU_abs{abs}_PCR_sub_filt.physeq \
OTU_abs{abs}_PCR_sub_filt_bray-NMDS.pdf
# DESeq2
SIPSimR phyloseq_DESeq2 \
--log2 {log2} \
--hypo greater \
--cont 1,3,5 \
--treat 2,4,6 \
OTU_abs{abs}_PCR_sub_filt.physeq \
> OTU_abs{abs}_PCR_sub_filt_DESeq2
In [ ]:
!chmod 777 $bashFileTmp
!cd $workDir; \
nestrun --template-file $bashFileTmp -d $buildDir --log-file HR-SIP.log -j 10
In [ ]:
%pushnote HR-SIP complete: $buildDir
In [ ]:
bashFileTmp = os.path.splitext(bashFile)[0] + '_qSIP.sh'
bashFileTmp
In [ ]:
%%writefile $bashFileTmp
#!/bin/bash
# qSIP
SIPSim qSIP \
OTU_abs{abs}.txt \
OTU_abs{abs}_PCR_sub.txt \
> OTU_abs{abs}_PCR_sub_qSIP.txt
# qSIP: atom excess
SIPSim qSIP_atomExcess \
--np {np} \
OTU_abs{abs}_PCR_sub_qSIP.txt \
{exp_design} \
> OTU_abs{abs}_PCR_sub_qSIP_atom.txt
In [ ]:
!chmod 777 $bashFileTmp
!cd $workDir; \
nestrun --template-file $bashFileTmp -d $buildDir --log-file qSIP.log -j 2
In [ ]:
%pushnote qSIP complete: $buildDir
In [ ]:
bashFileTmp = os.path.splitext(bashFile)[0] + '_dBD.sh'
bashFileTmp
In [ ]:
%%writefile $bashFileTmp
#!/bin/bash
#deltaBD
SIPSim deltaBD \
OTU_abs{abs}_PCR_sub.txt \
{exp_design} \
> OTU_abs{abs}_PCR_sub_dBD.txt
In [ ]:
!chmod 777 $bashFileTmp
!cd $workDir; \
nestrun --template-file $bashFileTmp -d $buildDir --log-file deltaBD.log -j 4
In [ ]:
%pushnote deltaBD complete: $buildDir
In [ ]:
bashFileTmp = os.path.splitext(bashFile)[0] + '_cMtx.sh'
bashFileTmp
In [ ]:
%%writefile $bashFileTmp
#!/bin/bash
# HR-SIP
SIPSimR DESeq2_confuseMtx \
--libs 2,4,6 \
--padj {padj} \
BD-shift_stats.txt \
OTU_abs{abs}_PCR_sub_filt_DESeq2
# HR-SIP multiple 'heavy' BD windows
SIPSimR DESeq2_confuseMtx \
--libs 2,4,6 \
--padj {padj} \
-o DESeq2_multi-cMtx \
BD-shift_stats.txt \
OTU_abs1e9_PCR_sub_BD3_DESeq2
# qSIP
SIPSimR qSIP_confuseMtx \
--libs 2,4,6 \
BD-shift_stats.txt \
OTU_abs{abs}_PCR_sub_qSIP_atom.txt
# heavy-SIP
SIPSimR heavy_confuseMtx \
--libs 2,4,6 \
BD-shift_stats.txt \
OTU_abs{abs}_PCR_sub.txt
In [ ]:
!chmod 777 $bashFileTmp
!cd $workDir; \
nestrun --template-file $bashFileTmp -d $buildDir --log-file cMtx.log -j 10
In [ ]:
def agg_cMtx(prefix):
# all data
#!nestagg delim \
# -d $buildDir \
# -k percIncorp,percTaxa,rep \
# -o $prefix-cMtx_data.txt \
# --tab \
# $prefix-cMtx_data.txt
# overall
x = prefix + '-cMtx_overall.txt'
!nestagg delim \
-d $buildDir \
-k shared_perc,perm_perc,rep \
-o $x \
--tab \
$x
# by class
x = prefix + '-cMtx_byClass.txt'
!nestagg delim \
-d $buildDir \
-k shared_perc,perm_perc,rep \
-o $x \
--tab \
$x
agg_cMtx('DESeq2')
agg_cMtx('qSIP')
agg_cMtx('heavy')
In [ ]:
%pushnote microBetaDiv complete!
In [ ]:
F = os.path.join(buildDir, '*-cMtx_byClass.txt')
files = glob.glob(F)
files
In [ ]:
%%R -i files
df_byClass = list()
for (f in files){
ff = strsplit(f, '/') %>% unlist
fff = ff[length(ff)]
df_byClass[[fff]] = read.delim(f, sep='\t')
}
df_byClass = do.call(rbind, df_byClass)
df_byClass$file = gsub('\\.[0-9]+$', '', rownames(df_byClass))
df_byClass$method = gsub('-.+', '', df_byClass$file)
rownames(df_byClass) = 1:nrow(df_byClass)
df_byClass %>% head(n=3)
In [ ]:
%%R
# renaming method
rename = data.frame(method = c('DESeq2', 'DESeq2_multi', 'heavy', 'qSIP'),
method_new = c('HR-SIP', 'MW-HR-SIP', 'Heavy-SIP', 'qSIP'))
df_byClass = inner_join(df_byClass, rename, c('method'='method')) %>%
select(-method) %>%
rename('method' = method_new)
df_byClass %>% head(n=3)
In [ ]:
%%R -w 800 -h 550
# summarize by SIPSim rep & library rep
df_byClass.s = df_byClass %>%
group_by(method, shared_perc, perm_perc, variables) %>%
summarize(mean_value = mean(values),
sd_value = sd(values))
# plotting
ggplot(df_byClass.s, aes(variables, mean_value, color=method,
ymin=mean_value-sd_value,
ymax=mean_value+sd_value)) +
geom_pointrange(alpha=0.8, size=0.2) +
labs(y='Value') +
facet_grid(shared_perc ~ perm_perc) +
theme_bw() +
theme(
text = element_text(size=16),
axis.title.x = element_blank(),
axis.text.x = element_text(angle=45, hjust=1)
)
In [ ]:
%%R -w 800 -h 600
# summarize by SIPSim rep & library rep
vars = c('Balanced Accuracy', 'Sensitivity', 'Specificity')
df_byClass.s.f = df_byClass.s %>%
filter(variables %in% vars)
# plotting
ggplot(df_byClass.s.f, aes(variables, mean_value, fill=method,
ymin=mean_value-sd_value,
ymax=mean_value+sd_value)) +
#geom_pointrange(alpha=0.8, size=0.2) +
geom_bar(stat='identity', position='dodge', width=0.8) +
geom_errorbar(stat='identity', position='dodge', width=0.8) +
scale_y_continuous(breaks=seq(0, 1, 0.2)) +
labs(y='Value') +
facet_grid(shared_perc ~ perm_perc) +
theme_bw() +
theme(
text = element_text(size=16),
axis.title.x = element_blank(),
axis.text.x = element_text(angle=50, hjust=1)
)
In [ ]:
%%R -w 800 -h 450
# summarize by SIPSim rep & library rep
vars = c('Balanced Accuracy', 'Sensitivity', 'Specificity')
df_byClass.s.f = df_byClass.s %>%
filter(variables %in% vars) %>%
ungroup() %>%
mutate(perm_perc = perm_perc %>% as.character,
perm_perc = perm_perc %>% reorder(perm_perc %>% as.numeric))
# plotting
ggplot(df_byClass.s.f, aes(shared_perc, mean_value,
color=perm_perc,
ymin=mean_value-sd_value,
ymax=mean_value+sd_value)) +
geom_pointrange(alpha=0.8, size=0.2) +
geom_line() +
scale_color_discrete('% of rank\nabundances\npermuted') +
labs(x='% taxa shared among replicate pre-fraction communities') +
facet_grid(variables ~ method, scales='free_y') +
theme_bw() +
theme(
text = element_text(size=16),
axis.title.y = element_blank()
)
In [60]:
atomX_files = glob.glob('*/*/*/*_atom.txt')
len(atomX_files)
Out[60]:
In [61]:
%%R -i atomX_files
df_atomX = list()
for(F in atomX_files){
tmp = read.delim(F, sep='\t')
FF = strsplit(F, '/') %>% unlist
tmp$shared_perc = FF[1]
tmp$perm_perc = FF[2]
tmp$rep = FF[3]
tmp$file = FF[4]
df_atomX[[F]] = tmp
}
df_atomX = do.call(rbind, df_atomX)
rownames(df_atomX) = 1:nrow(df_atomX)
df_atomX %>% head(n=3)
In [62]:
BDshift_files = glob.glob('*/*/*/BD-shift_stats.txt')
len(BDshift_files)
Out[62]:
In [63]:
%%R -i BDshift_files
df_shift = list()
for(F in BDshift_files){
tmp = read.delim(F, sep='\t')
FF = strsplit(F, '/') %>% unlist
tmp$shared_perc = FF[1]
tmp$perm_perc = FF[2]
tmp$rep = FF[3]
tmp$file = FF[4]
df_shift[[F]] = tmp
}
df_shift = do.call(rbind, df_shift)
rownames(df_shift) = 1:nrow(df_shift)
df_shift = df_shift %>%
filter(library %in% c(2,4,6)) %>%
group_by(taxon, shared_perc, perm_perc, rep) %>%
summarize(median = median(median)) %>%
ungroup() %>%
rename('median_true_BD_shift' = median)
df_shift %>% head(n=3)
In [64]:
%%R
df.j = inner_join(df_atomX, df_shift, c('taxon' = 'taxon',
'shared_perc'='shared_perc',
'perm_perc'='perm_perc',
'rep'='rep')) %>%
filter(!is.na(BD_diff)) %>%
mutate(true_incorporator = ifelse(median_true_BD_shift > 0.002, TRUE, FALSE),
true_atom_fraction_excess = median_true_BD_shift / 0.036,
atom_fraction_excess = ifelse(is.na(atom_CI_low), 0, atom_fraction_excess))
df.j %>% head(n=3)
In [ ]:
%%R
# free memory
df_shift = df_atomX = NULL
In [65]:
%%R -h 250
# difference between true and estimated
df.j.dis.qSIP = df.j %>%
mutate(delta_excess = atom_fraction_excess * 100 - true_atom_fraction_excess * 100) %>%
group_by(shared_perc, perm_perc, true_incorporator) %>%
summarize(mean_delta_excess = mean(delta_excess),
sd_delta_excess = sd(delta_excess)) %>%
ungroup() %>%
mutate(shared_perc = shared_perc %>% reorder(shared_perc %>% as.numeric),
perm_perc = perm_perc %>% reorder(perm_perc %>% as.numeric)) %>%
filter(true_incorporator == TRUE)
# plotting
ggplot(df.j.dis.qSIP, aes(shared_perc, mean_delta_excess,
color=perm_perc, group=perm_perc,
ymin=mean_delta_excess-sd_delta_excess,
ymax=mean_delta_excess+sd_delta_excess)) +
geom_point() +
geom_linerange(alpha=0.5, size=0.8) +
geom_line() +
scale_color_discrete('% taxa ranks\npermuted') +
labs(x='% taxa shared', y='atom % excess\n(truth - estimate)') +
theme_bw() +
theme(
text = element_text(size=16)
)
In [66]:
dBD_files = glob.glob('*/*/*/*_dBD.txt')
len(dBD_files)
Out[66]:
In [67]:
%%R -i dBD_files
df_dBD = list()
for(F in dBD_files){
tmp = read.delim(F, sep='\t')
FF = strsplit(F, '/') %>% unlist
tmp$shared_perc = FF[1]
tmp$perm_perc = FF[2]
tmp$rep = FF[3]
tmp$file = FF[4]
df_dBD[[F]] = tmp
}
df_dBD = do.call(rbind, df_dBD)
rownames(df_dBD) = 1:nrow(df_dBD)
df_dBD %>% head(n=3)
In [ ]:
BDshift_files = glob.glob('*/*/*/BD-shift_stats.txt')
len(BDshift_files)
Out[ ]:
In [ ]:
%%R -i BDshift_files
df_shift = list()
for(F in BDshift_files){
tmp = read.delim(F, sep='\t')
FF = strsplit(F, '/') %>% unlist
tmp$shared_perc = FF[1]
tmp$perm_perc = FF[2]
tmp$rep = FF[3]
tmp$file = FF[4]
df_shift[[F]] = tmp
}
df_shift = do.call(rbind, df_shift)
rownames(df_shift) = 1:nrow(df_shift)
df_shift = df_shift %>%
filter(library %in% c(2,4,6)) %>%
group_by(taxon, shared_perc, perm_perc, rep) %>%
summarize(median = median(median)) %>%
ungroup() %>%
rename('median_true_BD_shift' = median)
df_shift %>% head(n=3)
In [ ]:
%%R
df.j = inner_join(df_dBD, df_shift, c('taxon' = 'taxon',
'shared_perc'='shared_perc',
'perm_perc'='perm_perc',
'rep'='rep')) %>%
filter(!is.na(delta_BD)) %>%
mutate(true_incorporator = ifelse(median_true_BD_shift > 0.002, TRUE, FALSE),
true_atom_fraction_excess = median_true_BD_shift / 0.036,
atom_fraction_excess = delta_BD / 0.036)
df.j %>% head(n=3)
In [ ]:
%%R
# free memory
df_shift = df_dBD = NULL
In [ ]:
%%R -h 250
# difference between true and estimated
df.j.dis.dBD = df.j %>%
mutate(delta_excess = atom_fraction_excess * 100 - true_atom_fraction_excess * 100) %>%
group_by(shared_perc, perm_perc, true_incorporator) %>%
summarize(mean_delta_excess = mean(delta_excess),
sd_delta_excess = sd(delta_excess)) %>%
ungroup() %>%
mutate(shared_perc = shared_perc %>% reorder(shared_perc %>% as.numeric),
perm_perc = perm_perc %>% reorder(perm_perc %>% as.numeric)) %>%
filter(true_incorporator == TRUE)
# plotting
ggplot(df.j.dis.dBD, aes(shared_perc, mean_delta_excess,
color=perm_perc, group=perm_perc,
ymin=mean_delta_excess-sd_delta_excess,
ymax=mean_delta_excess+sd_delta_excess)) +
geom_point() +
geom_linerange(alpha=0.5, size=0.8) +
geom_line() +
scale_color_discrete('% taxa ranks\npermuted') +
labs(x='% taxa shared', y='atom % excess\n(truth - estimate)') +
theme_bw() +
theme(
text = element_text(size=16)
)
In [ ]:
%%R -w 670 -h 250
df.jj = rbind(df.j.dis.qSIP %>% mutate(method='qSIP'),
df.j.dis.dBD %>% mutate(method='ΔBD'))
# plotting
ggplot(df.jj, aes(shared_perc, mean_delta_excess,
color=perm_perc, group=perm_perc,
ymin=mean_delta_excess-sd_delta_excess,
ymax=mean_delta_excess+sd_delta_excess)) +
geom_point() +
geom_linerange(alpha=0.5, size=0.5) +
geom_line() +
geom_line() +
scale_color_discrete('% taxa ranks\npermuted') +
labs(x='% taxa shared', y='13C atom % excess\n(truth - estimate)') +
facet_grid(. ~ method) +
theme_bw() +
theme(
text = element_text(size=16)
)
In [ ]: