Rob’s Code

Big batching your RNAseq

2020-10-04T00:00:00+00:00

So did you do read alignment or pseudoalignment? Why not both?

Everyone will have an opinion on whether psuedoalignment is a good thing, so if you are going to use it, at least one collaborator will inevitably suggest you go back and try it the “other” way. Head them off and do both, since pseudoalignment costs very little compute.

#!/usr/bin/env bash
exec 1>> command.log 2>&1
set -ex

##################################
# for running all of the trimming, alignment and alignment-free pipelines
# does not make use of unpaired outputs
# this script is set for a server with 100 threads and 256GB memory
##################################
# directory where everything will go
BASE_DIR=/path/to/base_dir
# directory of reads must have *_R1_001.fastq.gz and *_R2_001.fastq.gz filenames
READ_DIR=${BASE_DIR}/read_folder
# prior to running, the appropriate reference files must be generated in refs:
# an assembly gtf (must match version that built indexes below):
GTF=${BASE_DIR}/refs/Rattus_norvegicus.Rnor_6.0.96.gtf
# a salmon index:
SALMON_INDEX=${BASE_DIR}/refs/Rnor6.quasi_index_21
# a STAR index:
STAR_INDEX=${BASE_DIR}/refs/Rnor6_star
# there is a script in refs, modify for correct species and assembly

to trim or not to trim

With better and better sequence quality, this is making less of a difference, but if for instance you have custom adapters, probably a good idea. trimmomatic is also good, but why deal with java when fastp is pretty fast.

##################################
# trim reads
mkdir ${BASE_DIR}/trimmed
cd ${BASE_DIR}/trimmed
# use paired end reads from read_dir
# trying fastp, uses 2 worker threads per run (approx 250%-400% cpu)
# for quality trimming, use -5 and -3 with default window(4) and mean q(20)
trim(){
  out="$(basename -s '_R1_001.fastq.gz' $1)"
  R2="$(echo $1 | sed s/_R1_001.fastq.gz/_R2_001.fastq.gz/)"
  fastp \
    --thread 2\
    -i $1 \
    -I $R2 \
    -o ${out}_R1_trim.fastq.gz \
    -O ${out}_R2_trim.fastq.gz \
    --adapter_sequence AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCT \
    --adapter_sequence_r2 GATCGGAAGAGCACACGTCTGAACTCCAGTCACATCACGATCTCGTATGCCGTCTTCTGCTTG \
    -3 \
    -5 \
    -l 20 \
    -h ${out}.html \
    -j ${out}.json
}

export -f trim
parallel -j50 trim {} ::: ${READ_DIR}/*R1_001.fastq.gz

salmon

#################################
# Salmon pipeline
mkdir ${BASE_DIR}/salmon
cd ${BASE_DIR}/salmon
# for paired end data
# salmon counts quasi-mapping task
salmon_map(){
  out="$(basename -s '_R1_trim.fastq.gz' $1)"
  R2="$(echo $1 | sed 's/_R1_trim/_R2_trim/')"
  salmon --no-version-check quant \
    --seqBias \
    --gcBias \
    --validateMappings \
    -i $2 \
    -l A \
    -p 20 \
    -1 <(pigz -dc "$1") \
    -2 <(pigz -dc "$R2") \
    -o ${out}
}

# run loop
export -f salmon_map
parallel -j5 salmon_map {} $SALMON_INDEX \
  ::: ${BASE_DIR}/trimmed/*_R1_trim.fastq.gz

STAR using shared memory

An advantage for using STAR is that it allows you to put the reference into shared memory and run several samples without eating up a lot of memory. Note the genomeLoad options

#################################
# STAR-featureCounts pipeline
mkdir ${BASE_DIR}/bam
cd ${BASE_DIR}/bam
# for paired end data
# STAR mapping task
star_map(){
  out="$(basename -s '_R1_trim.fastq.gz' $1)"
  R2="$(echo $1 | sed 's/_R1_trim/_R2_trim/')"
  STAR \
    --runThreadN 20 \
    --genomeDir $2 \
    --genomeLoad LoadAndKeep \
    --readFilesCommand zcat \
    --outFileNamePrefix ${out}_star \
    --outSAMtype BAM SortedByCoordinate \
    --outSAMmapqUnique 60 \
    --outSAMattrRGline ID:$out LB:lib1 PL:illumina PU:unit1 SM:$out \
    --outBAMsortingThreadN 10 \
    --outBAMsortingBinsN 10 \
    --limitBAMsortRAM 10000000000 \
    --outBAMcompression 10 \
    --bamRemoveDuplicatesType UniqueIdentical \
    --outFilterIntronMotifs RemoveNoncanonical \
    --outFilterMismatchNmax 2 \
    --outFilterScoreMinOverLread 0.30 \
    --outFilterMatchNminOverLread 0.30 \
    --alignSoftClipAtReferenceEnds No\
    --readFilesIn $1 $R2
}

# add to shared memory
STAR \
  --runThreadN 100 \
  --genomeDir $STAR_INDEX \
  --genomeLoad LoadAndExit

# run loop
export -f star_map
parallel -j5 star_map {} $STAR_INDEX ::: ${BASE_DIR}/trimmed/*_R1_trim.fastq.gz

# remove from shared memory
STAR \
  --genomeDir $STAR_INDEX \
  --genomeLoad Remove

# featureCounts for all in folder (max threads 64)
FC_OUTFILE="featureCounts"
featureCounts \
  -T 64 \
  -p \
  -B \
  -C \
  -D 2000 \
  -t exon \
  -g gene_id \
  -a $GTF \
  -o ${FC_OUTFILE}_counts.txt *.bam

########################################
# go home
cd ${BASE_DIR}

Next time

There it is, that script with 100 threads and 256GB memory can run a hundred samples easily overnight. In the future I can speed test the actual throughput.

Deploying R in conda

2020-09-24T00:00:00+00:00

a solution to the problem of portability

Working with R, you may have come across the frustration of figuring out where are put stuff, or the more frustrating ‘trying to install R as user’. Conda has spent a fair but of time setting it up so that conda virtual environments can store all of the necessary files, and access local libraries very efficiently.

conda create -n r4.0 -y python=3 r-base r-essentials r-devtools

So now we are done right?

Not exactly, as a plus, you can specify your R version with r-base==3.6.3 and if you are wondering about bioconductor, then simply bioconductor-package will install any package in the right place. However, you will eventually run into the problem that there is an R package that isn’t available in conda. In this case, you can do as below, where I build a custom r-tigris and r-gg.gap package for a shiny app with maps. You can actually push these to your private conda channel in the cloud if you register for one, but more often, you will build it where you want it, and specify --use-local in your install to have conda check the locally built things. Also, you can clear out the temo build files with conda build purge

#!/usr/bin/env bash
exec 1>> dashboard_build.log 2>&1
set -ex

# construct tigris
conda skeleton cran tigris
conda-build r-tigris

# construct gg.gap
conda skeleton cran gg.gap
conda-build r-gg.gap

# cleanup after builds
conda-build purge
rm -rf r-tigris/
rm -rf r-gg.gap/

# build initial
conda create -n shiny -y --use-local python=3 r-tigris r-gg.gap r-shiny \
  r-readr r-tidyr r-dplyr r-ggplot2 r-data.table r-shinyalert \
  r-rcolorbrewer r-gridextra r-stringr r-shinydashboard r-ggpubr r-ggplotify \
  r-leaflet r-units r-sf r-rgdal r-leafsync r-reshape2

# # Can be launched from Rscript (run in background screen)
# screen -S covid
# source $(conda info --base)/etc/profile.d/conda.sh
# conda activate shiny || { echo "shiny Conda environment not activated"; exit; }
# Rscript covid_19_data_viz_v6.R

Next time

One of the things this doesn’t address is installing github packages. If they are simple, then hopefully after you are done installing R (include r-devtools), you can run R and install_github('', dependencies=FALSE) from inside. However, beware, because if it has dependencies, R will try to install them where it knows to: /usr/local and this will create some havoc. Instead, you may have to research those dependencies and install them prior with conda, as I did here with r-units, r-sf, r-rgdal. Then, go back into R and try again. Repeat until done. Ugh, never an easy answer is there.

Release the kraken(2)!

2020-09-23T00:00:00+00:00

…just a little bit

Getting kraken2 to be set up for both Braken and qiime2 takes a little bit of work. First, you need to have the standard report for Braken, and you need the --mpa-report to use parse_kraken_to_biom.py from this helpful analysis. In order to get that going, its best to start small and test your first 10 samples. My samples are not ordered, so the first 10 included a mix of all experimental groups. Check that, otherwise you may threshold incorrectly.

#!/usr/bin/env bash
exec 1>> command.log 2>&1
set -ex

####KRAKEN
kdb='/path/to/db/minikraken2_v2_8GB_201904_UPDATE'
mkdir -p kraken
mkdir -p kraken_test

# test run (species richness vs confidence)
parallel -j10 --plus --rpl '{qq} s:.*/::; s:_R1.fastq.gz::;' kraken2 \
  --threads 10 \
  --db $kdb \
  --paired \
  --confidence {2} \
  --report kraken_test/{1qq}_{2}.mpa_report\
  --use-mpa-style \
  --output kraken_test/{1qq}_{2}.kraken \
  {1} {1/_R1.f/_R2.f} \
  ::: $(awk 'FS="," {print $2}' 20200905_manifest.csv | head -n11) \
  ::: 0 0.05 0.1 0.15 0.2 0.25 0.3 0.4 0.5 0.6 0.7 0.8 0.9

# calculate richness graph for confidence threshold
wc -l kraken_test/*.mpa_report > richness_test.txt
python kraken_test.py

Note the range of confidence thresholds

You are looking for the elbow of the curve, but go too far right and you end up removing all your OTUs. Here is a much better thread on it. Well, lets see what I have:

#!/usr/bin/env python

import numpy as np
import seaborn as sns
import pandas as pd

# import data, split filenames into samples & confidence
a = pd.read_table('richness_test.txt', header=None, sep='\s+')
a[1] = a[1].str.replace("kraken_test/", "")
a[1] = a[1].str.replace(".mpa_report", "")
a[['sample', 'confidence']] = a[1].str.rsplit("_", 1, expand=True)
a.rename(columns={0:'richness'}, inplace=True)

# make plot
b = sns.lineplot(x='confidence', y='richness', data=a, hue='sample')
b.set_ylabel("Crude richness", size=14)
b.set_xlabel("Confidence threshold", size=12)
b.figure.savefig('richness_test.pdf')

Will give you a plot of

So, looks like 0.1 would be a good threshold to avoid noise. If you have too sparse data later on, you could dial it back closer to 0.05. So, lets break the server!

# with only 256GB of RAM, limit to 24 simultaneous instances(x4 = 96 threads)
parallel -j24 --plus --rpl '{qq} s:.*/::; s:_R1.fastq.gz::' kraken2 \
  --threads 4 \
  --db $kdb \
  --paired \
  --confidence 0.1 \
  --report kraken/{qq}.mpa_report \
  --use-mpa-style \
  --output kraken/{qq}.kraken \
  {} {/_R1.f/_R2.f} \
  ::: $(awk 'FS="," {print $2}' 20200905_manifest.csv)

# make standard report for braken
parallel -j24 --plus --rpl '{qq} s:.*/::; s:_R1.fastq.gz::' kraken2 \
  --threads 4 \
  --db $kdb \
  --paired \
  --confidence 0.1 \
  --report kraken/{qq}.report \
  --output kraken/{qq}.kraken \
  {} {/_R1.f/_R2.f} \
  ::: $(awk 'FS="," {print $2}' 20200905_manifest.csv)

# calculate richness graph for rough estimate
wc -l kraken/*.mpa_report > richness.txt
python kraken_richness.py

Lets do a quick check

#!/usr/bin/env python

import numpy as np
import seaborn as sns
import pandas as pd
import matplotlib.pyplot as plt

# import raw richness data
a = pd.read_table('richness.txt', header=None, sep='\s+')
a[1] = a[1].str.replace("kraken/", "")
a[1] = a[1].str.replace(".mpa_report", "")
a.rename(columns={0:'richness', 1:'specimen'}, inplace=True)

# merge with manifest
b = pd.read_csv('20200905_manifest.csv')
b = b.merge(a)

# plot raw richness (with four variables)
fig, axs = plt.subplots(nrows=1, ncols=4, figsize=(16,8))
blist = ['braaksc', 'ceradsc', 'cerIO', 'cogIO']

for i in range(4):
    sns.stripplot(x=blist[i], y='richness', data=b, hue='apoe_genotype', 
        dodge=True, alpha=0.5, zorder=0.5, ax=axs[i])
    sns.boxplot(x=blist[i], y='richness', data=b, boxprops={'facecolor':'None'}, 
        showfliers=False, ax=axs[i]).set_ylabel('Crude richness', size=12)

# fig.show()
fig.tight_layout()
fig.savefig('richness_raw.pdf')

Next time

That worked out pretty well, the plot at the top is the crude result. I have much more analysis to do, but something has caught my eye. Next up lets get that data in biom format with parse_kraken_to_biom.py

Installing geneshot (via conda)

2020-09-22T00:00:00+00:00

Installing geneshot (via conda)

So sometimes you can’t use docker (it happens to the best of us). In some cases, like high security environments. Geneshot is one of those neat looking new things I have been itching to try out. Luckily, miniconda is pretty flexible these days. Use an install script like:

#!/usr/bin/env bash
exec 1>> command.log 2>&1
set -ex

# build conda environment (r-breakaway built for 3.6.3)
conda create -n geneshot -y -vv python=3 nextflow r-tidyverse r-devtools \
  r-vroom bioconductor-phyloseq cutadapt bwa megahit prodigal metaphlan2 diamond \
  biopython mmseqs2==7.4e23d zarr pytables pyarrow
# --max-seqs option dropped in mmseqs2 >7
# pyarrow downgrades diamond quite a bit

# activate
source $(conda info --base)/etc/profile.d/conda.sh
conda activate geneshot || { echo "geneshot Conda environment not activated"; exit; }

# install non-conda components (required older python-dateutil for famli)
pip install barcodecop
pip install git+https://github.com/FredHutch/FAMLI.git@v1.5
pip install git+https://github.com/FredHutch/find-cags.git@v0.13.0

git clone https://github.com/Golob-Minot/geneshot.git -b v0.8.6
# need some fastatools to run, put them in the conda env bin folder
git clone https://github.com/Golob-Minot/fastatools.git
chmod a+x fastatools/*.py
for k in fastatools/*.py
  do
    j=$(basename $k)
    cp $k $(conda info --base)/envs/geneshot/bin/$j
done
rm -rf fastatools

# need github R packages, for now install with R script (fix later)
#####R code
#devtools::install_github(c("adw96/breakaway", "bryandmartin/corncob"), upgrade="never")
#####
echo 'devtools::install_github(c("adw96/breakaway", "bryandmartin/corncob"), upgrade="never")' \
  > devtools.R
Rscript devtools.R
rm devtools.R

Did it work?

As far as testing goes, these seem to be good to go mostly, but I get some issues running corncob. Luckily, that can be run separately with run_corncob.nf

##PASS w/o formula
NXF_VER=20.04.1 nextflow run main.nf \
    -c nextflow.config \
    -profile testing \
    --manifest data/mock.manifest.csv \
    --nopreprocess \
    --output output \
    --hg_index data/hg_chr_21_bwa_index.tar.gz \
    --distance_threshold 0.5 \
    # -w work/ \
    # --noannot \
    # --formula "label1 + label2" \

##FAIL
NXF_VER=20.04.1 nextflow run main.nf \
    -c nextflow.config \
    -profile testing \
    --manifest data/mock.manifest.csv \
    --preprocess_output output/preprocess_output \
    --output output1 \
    --hg_index data/hg_chr_21_bwa_index.tar.gz \
    --formula "label1 + label2" \
    --distance_threshold 0.1 \
    -w work/ \
    --noannot \
    --savereads \
    -resume

# PASS # Test with preprocessing and no formula
NXF_VER=20.04.1 nextflow run main.nf \
    -c nextflow.config \
    -profile testing \
    --manifest data/mock.manifest.csv \
    --preprocess_output output/preprocess_output \
    --output output2 \
    --hg_index data/hg_chr_21_bwa_index.tar.gz \
    --distance_threshold 0.1 \
    -w work/ \
    --noannot \
    -resume

# FAIL # Test with formula and no preprocessing
NXF_VER=20.04.1 nextflow run main.nf \
    -c nextflow.config \
    -profile testing \
    --nopreprocess \
    --manifest data/mock.manifest.csv \
    --output output3 \
    --hg_index data/hg_chr_21_bwa_index.tar.gz \
    --formula "label1 + label2" \
    --distance_threshold 0.1 \
    -w work/ \
    --noannot \
    -resume

# PASS # Test with no formula and no preprocessing
NXF_VER=20.04.1 nextflow run main.nf \
    -c nextflow.config \
    -profile testing \
    --nopreprocess \
    --manifest data/mock.manifest.csv \
    --output output4 \
    --hg_index data/hg_chr_21_bwa_index.tar.gz \
    -w work/ \
    --noannot \
    -resume

# PASS # Test with the gene catalog made in a previous round
NXF_VER=20.04.1 nextflow run main.nf \
    -c nextflow.config \
    -profile testing \
    --gene_fasta data/genes.fasta.2.gz \
    --nopreprocess \
    --manifest data/mock.manifest.csv \
    --output output5 \
    --hg_index data/hg_chr_21_bwa_index.tar.gz \
    -w work/ \
    --noannot \
    -resume

So, always good to check

Looking at the head of your data is always good, and mine runs thus far. Now to let it run through 600 samples. Even with 100 threads + 360GB memory, that will take a little bit…

# geneshot test run
head -n15 20200905_manifest.csv > test_manifest.csv
nextflow \
    run \
        geneshot \
        --manifest test_manifest.csv \
        --output test \
        --nopreprocess \
        -with-report \
        -resume

Next time

Its worth noting, that -resume doesn’t work very well for me, it often starts over and I can’t figure out why. Also, my reads were already cleaned prior to running that, so I used --nopreprocess. However preprocessing did work in the test files, so it should be A-OK. In the future, time to solve the corncob part…

Allelic Imbalance of Expression in Rats: Part I

2020-09-21T00:00:00+00:00

Allelic imbalance of expression part I: getting your reads

But I can’t use WASP…

Sometimes you want to look at allelic imbalance of expression and WASP isn’t an option (like when you are looking at non-human genomes, or those for which you don’t have a personal genome). The fallback method to correct for mapping bias is N-masking the SNPs in the your samples, then mapping your samples to said masked fasta [Details].

First things first, you probably already have bams from your [DE analysis]. Take a subset of your sample bams, evenly distributed across your groups and merge it into a large bam:

# list of bams is two each of three tissues for A and B (12 total)
# merge bams
samtools merge -f -@ $threads -b $bams_to_merge $merge_bam
samtools index -@ $threads $merge_bam

# merge calls
gatk --java-options "-Xmx300G" HaplotypeCaller \
  -R $genome_ref \
  -D $dbSNP \
  --native-pair-hmm-threads $threads \
  -stand-call-conf 30 \
  -I $merge_bam \
  -O $merge_vcf

Then you can take these merged calls and mask your genome. If you are unfamiliar with loading STAR into shared memory, check out the mapping post. Then remap:

#######################################
# masking genome
bedtools maskfasta -fi $genome_ref -bed $merge_vcf -fo $mask_fa
samtools faidx $mask_fa

#######################################
# remapping to masked fasta

# building masked STAR index
mkdir -p $STAR_INDEX
STAR \
  --runMode genomeGenerate \
  --runThreadN $threads \
  --genomeDir $STAR_INDEX \
  --genomeFastaFiles $mask_fa \
  --sjdbGTFfile $GTF

# STAR-featureCounts pipeline
mkdir -p masked_bams
cd masked_bams

# add index to shared memory
STAR \
  --runThreadN $threads \
  --genomeDir $STAR_INDEX \
  --genomeLoad LoadAndExit

# run loop
parallel -j5 --plus \
  --rpl '{..} s:_R1_trim\.fastq\.gz::;s:/data/butlerr/nicotine_sensi/trimmed/::' \
  STAR \
    --runThreadN 20 \
    --genomeDir $STAR_INDEX \
    --genomeLoad LoadAndKeep \
    --readFilesCommand zcat \
    --outFileNamePrefix {..} \
    --outSAMtype BAM SortedByCoordinate \
    --outSAMmapqUnique 60 \
    --outSAMattrRGline ID:{..} LB:lib1 PL:illumina PU:unit1 SM:{..} \
    --outBAMsortingThreadN 10 \
    --outBAMsortingBinsN 10 \
    --limitBAMsortRAM 10000000000 \
    --outBAMcompression 10 \
    --bamRemoveDuplicatesType UniqueIdentical \
    --outFilterIntronMotifs RemoveNoncanonical \
    --outFilterMismatchNmax 2 \
    --outFilterScoreMinOverLread 0.30 \
    --outFilterMatchNminOverLread 0.30 \
    --alignSoftClipAtReferenceEnds No \
    --readFilesIn {} {/_R1_/_R2_} \
  :::: $fasta_to_read

# remove from shared memory
STAR \
  --genomeDir $STAR_INDEX \
  --genomeLoad Remove

# indexing bams
parallel -j $thr_para samtools index {} ::: *.bam
cd $wk_dir 
# you could do ../, but better to explicitly define your directories at the top
# of the script

Ok, now to do it all again

So, with your reads aligned with less mapping reference bias. STAR has implemented the WASP algorithm, but that is something for me to try in the future. Something that can save lots of time is gatk4.1 -ERC GVCF option. This will give you a set of gvcfs that can be combined with GenotypeGVCFs to give you an ‘all calls’ vcf with each sample called at all positions. So now you have read counts for REF and ALT everywhere (Note: I have three tissues I am looking at, so I split files up by a specific sample name pattern, then run the resulting steps in parallel -j3):

# calling gvcfs from masked bams
mkdir -p masked_gvcfs
parallel -j $thr_para --plus gatk --java-options \'"-Xmx4G"\' HaplotypeCaller \
  -R $genome_ref \
  -D $dbSNP \
  -stand-call-conf 30 \
  -I {} \
  -O masked_gvcfs/{/...}${suffix2} \
  -ERC GVCF \
  ::: $(ls masked_bams/*.bam)

cd masked_gvcfs

# tissue specific list files
ls -v *-1*Aligned_masked.g.vcf.gz | sed -r 's/(.*)Aligned/\1\t\1Aligned/' > vta.txt
ls -v *-2*Aligned_masked.g.vcf.gz | sed -r 's/(.*)Aligned/\1\t\1Aligned/' > nac.txt
ls -v *-3*Aligned_masked.g.vcf.gz | sed -r 's/(.*)Aligned/\1\t\1Aligned/' > nash.txt

# make list bed for reference genome (without patches)
awk 'BEGIN {FS="\t"}; {print $1 FS "0" FS $2}' $genome_ref.fai \
  | grep '^[1-9XY]' \
  > chr_list.bed

# make tissue specific gdbs (makes use of all 100 threads)
mkdir -p /data/butlerr/tmp
parallel -j3 gatk --java-options \'"-Xmx100G -Xms100G"\' GenomicsDBImport \
  --genomicsdb-workspace-path {} \
  --sample-name-map {}.txt \
  --max-num-intervals-to-import-in-parallel 24 \
  --intervals chr_list.bed \
  --tmp-dir=/data/butlerr/tmp \
  ::: "vta" "nac" "nash"

# call Joint Genotypes
parallel -j3 gatk --java-options \'"-Xmx40G"\' GenotypeGVCFs \
  -R $genome_ref \
  -D $dbSNP \
  -V gendb://{} \
  -L $BED \
  --merge-input-intervals true \
  --tmp-dir=/data/butlerr/tmp \
  -O {}_output.vcf.gz \
  ::: "vta" "nac" "nash"

One last issue though, I cannot get gatk to properly call only snps in the specified --intervals region, it just seems to ignore the option. Oh well, bcftools can get you there.

# won't filter based on --interval/-L try bcftools
parallel -j3 --plus bcftools query \
  -H \
  -R $BED \
  -f\''%CHROM\t%POS\t%REF\t%ALT\t%ID[\t%AD\t%DP]\n'\' \
  -o {..}.plot \
  {} \
  ::: *output.vcf.gz

Next time

What to do with all that read count data.

The Genetic Relevance of Human Induced Pluripotent Stem Cell-Derived Microglia to Alzheimer’s Disease and Major Neuropsychiatric Disorders

2019-07-23T00:00:00+00:00

The Genetic Relevance of Human Induced Pluripotent Stem Cell-Derived Microglia to Alzheimer’s Disease and Major Neuropsychiatric Disorders

Robert R Butler III, Alena Kozlova, Hanwen Zhang, Siwei Zhang, Michael Streit, Alan R Sanders, Krzysztof Laudanski, Zhiping P Pang, Pablo V Gejman, Jubao Duan

Journal: Molecular Neuropsychiatry
PMID: 32399472
DOI: 10.1159/000501935

Clinotator: analyzing ClinVar variation reports to prioritize reclassification efforts

2018-06-20T00:00:00+00:00

Clinotator: analyzing ClinVar variation reports to prioritize reclassification efforts

Robert R Butler III, Pablo V Gejman

Journal: F1000Research
PMID: 29862020
DOI: 10.12688/f1000research.14470.2

PhenotypeXpression: sub-classification of disease states using public gene expression data and literature

2018-01-01T00:00:00+00:00

PhenotypeXpression: sub-classification of disease states using public gene expression data and literature

Lucy Lu Wang, Huaiying Lin, Xiaojun Bao, Subhajit Sengupta, Ben Busby, Robert R Butler III

Journal: biorXiv
DOI: 10.1101/461301

Complete genome sequences of two human oral microbiome commensals: Streptococcus salivarius ATCC 25975 and S. salivarius ATCC 27945

2017-06-15T00:00:00+00:00

Complete genome sequences of two human oral microbiome commensals: Streptococcus salivarius ATCC 25975 and S. salivarius ATCC 27945

Robert R Butler, Jahna TA Soomer-James, Michel Frenette, Jean-François Pombert

Journal: Genome announcements
PMID: 28619805
DOI: 10.1128/genomeA.00536-17

Genetic characterization of the exceptionally high heat resistance of the non-toxic surrogate Clostridium sporogenes PA 3679

2017-04-03T00:00:00+00:00

Genetic characterization of the exceptionally high heat resistance of the non-toxic surrogate Clostridium sporogenes PA 3679

Robert R Butler III, Kristin M Schill, Yun Wang, Jean-François Pombert

Journal: Frontiers in Microbiology
PMID: 28421047
DOI: 10.3389/fmicb.2017.00545