This repository holds scripts and instructions for reproducing the benchmarking for neoepiscope as described in our manuscript.
bgzip v1.4 (from htslib)
HapCUT2 (we used commit 0f4ef9a0ea07d7eb32b306debee96c4ad00a13c2)
NeoPredPipe (we used commit 6a1ad97029c76ad16280a121d0ecc26768e3664e)
tabix v1.4 (from htslib)
Variant Effect Predictor (VEP) v91.3
Hg38 reference fasta (available from the Broad Institute resource bundle)
Hg38 DBSNP VCF (available from the Broad Institute resource bundle)
VEP GRCh38 cache (available from Ensembl)
We used WES paired-end fastq files from Bassani-Sternberg et al. [1] for benchmarking. The files are available at the European Genome-phenome Archive (EGA) under accession number EGAS00001002050. We used matched tumor and normal samples from patients Mel5, Mel8, Mel12, Mel15, and Mel16.
-
Run
neoepiscope download, answering yes to downloading/indexing the GENCODE v29 annotation, and yes to downloading the hg38 bowtie index. When asked if you would like to integrate NetMHCpan 4.0, answer yes, and provide the path to the NetMHCpan 4.0 directory you installed above. -
Move the VEP cache data to a subdirectory called 'cache' in your VEP main directory
-
Copy the 'chr_synonyms.txt' file in the VEP cache data to your VEP main directory
-
Move the VEP Downstream plugin to a subdirectory called 'plugins' in your VEP main directory
-
For each patient, make a config file for TSNAD. Using our template, replace 'INPUT_FILE_HERE' with the path to that patient's future TSNAD VEP output (e.g. /PATH/TO/THIS/REPOSITORY/Mel5_tumor_v_Mel5_normal.mutect.tumor.annotated.filtered.txt for Mel5). Replace OUTPUT_FOLDER_HERE with the patient-specific TSNAD output directory (e.g. /PATH/TO/THIS/REPOSITORY/Mel5_tsnad/ for Mel5). Replace NETMHC_HERE with the path to your netMHCpan 4.0 directory.
-
For each patient, make an HLA file for NeoPredPipe. Using our template, replace the two 'PATIENT' slots in the first column of the second row with the patient ID (e.g. Mel5).
We ran our benchmarking on an exclusive node of our institution's computer cluster, using the node's first four processors for multithreading (when possible). Each script benchmarks a different step in the neoepitope calling pipeline, and produces relevant output files from these steps along with files summarizing the CPU information and run time information (we used real time in our assessments). As an example, below are the commands to run the benchmarking for patient Mel5 from within the repository:
1) Benchmark BWA performance for tumor and normal samples:
SCRIPTS:
COMMANDS:
taskset -c 0,1,2,3 ./benchmark_bwa.sh HOME_DIRECTORY Mel5_tumor TUMOR_FASTQ1 TUMOR_FASTQ2 REFERENCE_FASTA BWA SAMTOOLS
taskset -c 0,1,2,3 ./benchmark_bwa.sh HOME_DIRECTORY Mel5_normal NORMAL_FASTQ1 NORMAL_FASTQ2 REFERENCE_FASTA BWA SAMTOOLS
INPUTS:
HOME_DIRECTORY is the path to this repository
TUMOR_FASTQ1/2 and NORMAL_FASTQ1/2 are the paired end tumor and normal fastqs
REFERENCE_FASTA is the path to your reference fasta
BWA is the path to your BWA executable
SAMTOOLS is the path to your samtools executable
OUTPUTS:
CPU info is output in Mel5_tumor.bwa.cpu_data and Mel5_normal.bwa.cpu_data
Run time info is output in Mel5_tumor.bwa.time_log and Mel5_normal.bwa.time_log
2) Benchmark BAM processing for tumor and normal samples
SCRIPTS:
benchmark_baserecalibration.sh
COMMANDS:
taskset -c 0,1,2,3 ./benchmark_markduplicates.sh HOME_DIRECTORY Mel5_tumor GATK
taskset -c 0,1,2,3 ./benchmark_markduplicates.sh HOME_DIRECTORY Mel5_normal GATK
taskset -c 0,1,2,3 ./benchmark_baserecalibration.sh HOME_DIRECTORY Mel5_tumor REFERENCE_FASTA DBSNP GATK PICARD
taskset -c 0,1,2,3 ./benchmark_baserecalibration.sh HOME_DIRECTORY Mel5_normal REFERENCE_FASTA DBSNP GATK PICARD
INPUTS:
HOME_DIRECTORY is the path to this repository
REFERENCE_FASTA is the path to your reference fasta
DBSNP is the path to your DBSNP VCF
GATK is the path to your GATK jar file
PICARD is the path to your PICARD executable
OUTPUTS:
CPU info is output in Mel5_tumor.markduplicates.cpu_data, Mel5_normal.markduplicates.cpu_data, Mel5_tumor.baserecalibration.cpu_data, and Mel5_normal.baserecalibration.cpu_data
Run time info is output in Mel5_tumor.markduplicates.time_log, Mel5_normal.markduplicates.time_log, Mel5_tumor.baserecalibration.time_log, and Mel5_normal.baserecalibration.time_log
3) Benchmark somatic variant calling
SCRIPTS:
COMMANDS:
taskset -c 0,1,2,3 ./benchmark_mutect.sh HOME_DIRECTORY Mel5_tumor Mel5_normal REFERENCE_FASTA DBSNP GATK
taskset -c 0,1,2,3 ./benchmark_filter_mutect.sh HOME_DIRECTORY Mel5_tumor Mel5_normal
INPUTS:
HOME_DIRECTORY is the path to this repository
REFERENCE_FASTA is the path to your reference fasta
DBSNP is the path to your DBSNP VCF
GATK is the path to your GATK jar file
OUTPUTS:
CPU info is output in Mel5_tumor_v_Mel5_normal.mutect.cpu_data, Mel5_tumor_v_Mel5_normal.mutect.cpu_data, Mel5_tumor_v_Mel5_normal.filtermutect.cpu_data and Mel5_tumor_v_Mel5_normal.filtermutect.cpu_data,
Run time info is output in Mel5_tumor_v_Mel5_normal.mutect.time_log, Mel5_tumor_v_Mel5_normal.mutect.time_log, Mel5_tumor_v_Mel5_normal.filtermutect.time_log, and Mel5_tumor_v_Mel5_normal.filtermutect.time_log
4) Benchmark germline variant calling
SCRIPTS:
COMMANDS:
taskset -c 0,1,2,3 ./benchmark_haplotypecaller.sh HOME_DIRECTORY Mel5_normal REFERENCE_FASTA DBSNP GATK
INPUTS:
HOME_DIRECTORY is the path to this repository
REFERENCE_FASTA is the path to your reference fasta
DBSNP is the path to your DBSNP VCF
GATK is the path to your GATK jar file
OUTPUTS:
CPU info is output in Mel5_normal.haplotypecaller.cpu_data
Run time info is output in Mel5_normal.haplotypecaller.time_log
5) Benchmark haplotype phasing for neoepiscope
SCRIPTS:
COMMANDS:
taskset -c 0,1,2,3 ./benchmark_hapcut2.sh HOME_DIRECTORY Mel5_tumor Mel5_normal HAPCUT2
INPUTS:
HOME_DIRECTORY is the path to this repository
HAPCUT2 is the path to your HapCUT2 build directory
OUTPUTS:
CPU info is output in Mel5_tumor_v_Mel5_normal.hapcut2.cpu_data
Run time info is output in Mel5_tumor_v_Mel5_normal.hapcut2.time_log
6) Benchmark HLA typing
SCRIPTS:
COMMANDS:
taskset -c 0,1,2,3 ./benchmark_optitype.sh HOME_DIRECTORY Mel5_tumor TUMOR_FASTQ1 TUMOR_FASTQ2 OPTITYPE CONFIG
INPUTS:
HOME_DIRECTORY is the path to this repository
TUMOR_FASTQ1/2 are the paired end tumor fastqs
OPTITYPE is the path to your Optitype python script
CONFIG is the path to your Optitype config file
OUTPUTS:
CPU info is output in Mel5_tumor.optitype.cpu_data
Run time info is output in Mel5_tumor.optitype.time_log
7) Benchmark pVACseq
SCRIPTS:
COMMANDS:
taskset -c 0,1,2,3 ./benchmark_vep.sh HOME_DIRECTORY Mel5_tumor Mel5_normal VEP_DIRECTORY GATK REFERENCE_FASTA REFERENCE_DICT PICARD BGZIP TABIX
taskset -c 0,1,2,3 ./benchmark_pvacseq.sh HOME_DIRECTORY Mel5_tumor Mel5_normal HLA-A*01:01,HLA-B*08:01,HLA-C*07:01
INPUTS:
HOME_DIRECTORY is the path to this repository
VEP_DIRECTORY is the path to your VEP directory
GATK is the path to your GATK jar file
REFERENCE_FASTA is the path to your reference fasta
REFERENCE_DICT is the path to your reference dictionary file
PICARD is the path to your Picard Tools executable
BGZIP is the path to your bgzip executable
TABIX is the path to your TABIX executable
OUTPUTS:
CPU info is output in Mel5_tumor_v_Mel5_normal.vep.cpu_data and Mel5_tumor_v_Mel5_normal.pvacseq.cpu_data
Run time info is output in Mel5_tumor_v_Mel5_normal.vep.time_log and Mel5_tumor_v_Mel5_normal.pvacseq.time_log
pVACseq neoepitopes are output in Mel5_tumor_v_Mel5_normal.final.tsv
8) Benchmark TSNAD
SCRIPTS:
COMMANDS:
taskset -c 0,1,2,3 ./benchmark_tsnad_vep.sh HOME_DIRECTORY Mel5_tumor Mel5_normal VEP_DIRECTORY
taskset -c 0,1,2,3 ./benchmark_tsnad.sh HOME_DIRECTORY Mel5 TSNAD CONFIG
INPUTS:
HOME_DIRECTORY is the path to this repository
VEP_DIRECTORY is the path to your VEP directory
TSNAD is the path to your TSNAD antigen_predicting_pipeline.py script
CONFIG is the path to your TSNAD config file for the patient (see above)
OUTPUTS:
CPU info is output in Mel5_tumor_v_Mel5_normal.vep_tsnad.cpu_data and Mel5_tumor_v_Mel5_normal.tsnad.cpu_data
Run time info is output in Mel5_tumor_v_Mel5_normal.vep_tsnad.time_log and Mel5_tumor_v_Mel5_normal.tsnad.time_log
TSNAD neoepitopes are output in Mel5_tsnad/predicted_neoantigen.txt
9) Benchmark MuPeXI
SCRIPTS:
COMMANDS:
taskset -c 0,1,2,3 ./benchmark_mupexi.sh HOME_DIRECTORY Mel5_tumor Mel5_normal HLA-A01:01,HLA-B08:01,HLA-C07:01 MUPEXI
INPUTS:
HOME_DIRECTORY is the path to this repository
MUPEXI is the path to your MuPeXI python script
OUTPUTS:
CPU info is output in Mel5_tumor_v_Mel5_normal.mupexi.cpu_data
Run time info is output in Mel5_tumor_v_Mel5_normal.mupexi.time_log
MuPeXI neoepitopes are output in Mel5_tumor_v_Mel5_normal.mupexi
10) Benchmark NeoPredPipe
SCRIPTS:
COMMANDS:
taskset -c 0,1,2,3 ./benchmark_neopredpipe.sh HOME_DIRECTORY Mel5_tumor Mel5_normal NEOPREDPIPE HLA_FILE
INPUTS:
HOME_DIRECTORY is the path to this repository
NEOPREDPIPE is the path to your NeoPredPipe main_netMHCpan_pipe.py script
HLA_FILE is the path to your NeoPredPipe HLA input file (see above)
OUTPUTS:
CPU info is output in Mel5_tumor_v_Mel5_normal.neopredpipe.cpu_data
Run time info is output in Mel5_tumor_v_Mel5_normal.neopredpipe.time_log
NeoPredPipe neoepitopes are output in Mel5_tumor_v_Mel5_normal.neoantigens.unfiltered.txt
11) Benchmark neoepiscope
SCRIPTS:
COMMANDS:
taskset -c 0,1,2,3 ./benchmark_neoepiscope.sh HOME_DIRECTORY Mel5_tumor Mel5_normal HLA-A*01:01,HLA-B*08:01,HLA-C*07:01
INPUTS:
HOME_DIRECTORY is the path to this repository
OUTPUTS:
CPU info is output in Mel5_tumor_v_Mel5_normal.neoepiscope.cpu_data
Run time info is output in Mel5_tumor_v_Mel5_normal.neoepiscope.time_log
NeoPredPipe neoepitopes are output in Mel5_tumor_v_Mel5_normal.neoepiscope.out and Mel5_tumor_v_Mel5_normal.neoantigens.Indels.unfiltered.txt
Post-processing
The same commands above can be run for patients Mel8, Mel12, Mel15, and Mel16 by replacing 'Mel5' with 'Mel8', 'Mel12', 'Mel15', or 'Mel16' in the commands.
To compile which epitope sequences were enumerated by each caller and for each patient, you can run our python script to parse the output of each tool for each patient.
python epitope_comparison.py -d HOME_DIRECTORY
INPUTS:
HOME_DIRECTORY is the path to this repository
OUTPUTS:
Mel5.peptide_overlap.out, Mel8.peptide_overlap.out, Mel12.peptide_overlap.out, Mel15.peptide_overlap.out, Mel16.peptide_overlap.out and combined.peptide_overlap.out are tab-delimited text files summarizing the neoepitopes predicted by all tools and which tools predicted them (on a per-patient basis, or for all patients combined).
Data availability
We used paired tumor-normal WES data of melanoma patients from Amaria et al., Carreno et al., Gao et al., Hugo et al., Roh et al., Snyder et al., Van Allen et al., and Zaretsky et al. (3-10); NSCLC patients from Rizvi et al. (11); and colon, endometrial, and thyroid cancer patients from Le et al. (12).
Read alignment and BAM processing
We aligned WES reads to the GRCh37d5 genome and generated genome-coordinate sorted alignments with duplicates marked using the Sanger cgpmap workflow (commit 0bacb0bee2e5c04b268c629d589ff1c551d34745). We realigned around indels and perform base recalibration using gatk-cocleaning-tool (commit d2bafc23221f6a8dceedd45a534163e0e1bf5c68). The relevant reference bundle is available here (see "Reference bundle"), and the necessary variant files can be found in the Broad Institute Resource Bundle.
The file fastq2bam.cwl.yaml can be used with sample_fastq2bam.sh and sample_fastq2bam.json to run the workflow. In the sample shell script and json file, change the paths to match the relevant paths for your computer.
Somatic variant calling
We used the mc3 workflow to call somatic variants. The relevant reference bundle is available here (see "Reference bundle"), and the necessary variant files can be found in the Broad Institute Resource Bundle.
The file mc3_variant.cwl (commit 72a24b55544e3011ede1c46b13d531a7d05ef4e0) can be used with sample_bam2variants.sh and sample_bam2variants.json to run the workflow. In the sample shell script and json file, change the paths to match the relevant paths for your computer.
Somatic variant processing/consensus calling
Following variant calling with mc3, we processed VCFs produced by MuSE, MuTect, Pindel, RADIA, SomaticSniper, and VarScan2 using vt:
First, we normalized each VCF:
vt normalize INPUT_VCF -n -r /PATH/TO/core_ref_GRCh37d5/genome.fa -o OUTPUT_VCF
Then we decomposed block substitutions:
vt decompose_blocksub -a -p INPUT_VCF -o OUTPUT_VCF
Then we decomposed multi-allelic variants:
vt decompose -s INPUT_VCF -o OUTPUT_VCF
Then we sorted variants:
vt sort INPUT_VCF -o OUTPUT_VCF
Then we took uniq variants:
vt uniq INPUT_VCF -o OUTPUT_VCF
After processing VCFs with vt, we ran VCF_parse.py to produce a consensus call set of variants produced by at least 2 callers and not overlapped by a Pindel variant, or called by Pindel:
python2.7 VCF_parse.py -v MuSE.uniq.vcf,MuTect.uniq.vcf,Pindel.uniq.vcf,RADIA.uniq.vcf,SomaticSniper.uniq.vcf,VarScan2.uniq.vcf -c MuSE,MuTect,Pindel,RADIA,SomaticSniper,VarScan2 -o /PATH/TO/OUTPUT_DIRECTORY -n 2 -s SAMPLE_NAME
This outputs a file called SAMPLE_NAME.consensus.vcf in the OUTPUT_DIRECTORY used for downstream analyses.
Germline variant calling and processing
We used GATK v3.7 to run HaplotypeCaller for calling germline variants, and to run VariantFiltration for filtering the results of HaplotypeCaller. The relevant reference bundle is available here (see "Reference bundle"), and the necessary variant files can be found in the Broad Institute Resource Bundle.
HaplotypeCaller was run using default options:
java -jar GenomeAnalysisTK.jar -R /PATH/TO/core_ref_GRCh37d5/genome.fa -T HaplotypeCaller -I /PATH/TO/NORMAL.realigned.cleaned.bam -o /PATH/TO/germline.vcf
VariantFiltration was run as below:
java -jar GenomeAnalysisTK.jar -R /PATH/TO/core_ref_GRCh37d5/genome.fa -T VariantFiltration --variant /PATH/TO/germline.vcf -o /PATH/TO/germline.filtered.vcf --clusterSize 3 --clusterWindowSize 15 --missingValuesInExpressionsShouldEvaluateAsFailing --filterName 'QDFilter' --filterExpression 'QD < 2.0' --filterName 'QUALFilter' --filterExpression 'QUAL < 100.0' --filterName DPFilter --filterExpression 'DP < 10.0'
Only variants passing all filters were retained for downstream analyses. We also ran vt to normalize, decompose, sort, and obtain a unique list of variants from the filtered germline VCFs (as described above for the VCFs from individual somatic variant callers).
Genomic coverage
We used bedtools (v2.23.0-19-ge65c98b) genomecov to determine the Mbp of genome covered in each tumor sample:
bedtools genomecov -ibam /PATH/TO/TUMOR_BAM -bg > /PATH/TO/OUTPUT/BEDGRAPH
We processed the output BedGraph file to determine the number of bp pairs that were covered by at least 3 reads (sufficient depth for somatic variant calling by SomaticSniper and VarScan 2). This data was stored in a tab-separated text file with columns for the patient ID, tumor ID, and Mbp covered (referenced as coverage_summary.tsv in post-processing R script).
Haplotype phasing
Before performing haplotype phasing, we swapped the sample columns in our consensus somatic variants VCF and merged our filtered germline and consensus somatic variants using neoepiscope:
neoepiscope swap -i SOMATIC_VCF -o SWAPPED_SOMATIC_VCF
neoepiscope merge -g GERMLINE_VCF -s SWAPPED_SOMATIC_VCF -o MERGED_VCF
Then, we predicted haplotypes using HapCUT2:
extractHAIRS --indels 1 --bam TUMOR_SAMPLE_NAME.realigned.cleaned.bam --VCF MERGED_VCF --out FRAGMENT_FILE
HAPCUT2 --fragments FRAGMENT_FILE --vcf MERGED_VCF --output HAPLOTYPES
Finally, we prepared our haplotype predictions for neoepiscope neoepitope prediction using neoepiscope:
neoepiscope prep -v MERGED_VCF -c HAPLOTYPES -o PREPPED_HAPLOTYPES
Neoepitope prediction
We predicted neoepitopes of 8-24 amino acids in length with neoepiscope, both accounting for phasing and germline and somatic variants, and not accounting for phasing:
neoepiscope call -b hg19 -c PREPPED_HAPLOTYPES -o OUTPUT_FILE -k 8,24
neoepiscope call -b hg19 -c PREPPED_HAPLOTYPES -o OUTPUT_FILE -k 8,24 --isolate --germline exclude
To determine the effects of phasing on neoepitope prediction, we used the script epitope_count_comparison.py to identify variants unique to/shared between the two neoepitope calling modes. The neoepiscope output files should be formatted as PATIENT_ID.TUMOR_ID.neoepiscope.out and PATIENT_ID.TUMOR_ID.neoepiscope.somatic_unphased.out for compatiblity with this script. The output file from this script (phasing_epitope_data.tsv) is processed by phasing_stats.R.
Phasing prevalence
We used a combination of two scripts to analyze phasing. First, we ran phasing_analysis.py to identify instances of variant phasing within 33, 72, or 94 bp across all patients and tumors:
python phasing_analysis.py -o OUTPUT_DIR -n NEOEPISCOPE_DATA_DIR -c HAPLOTYPE_DIR -d 33
python phasing_analysis.py -o OUTPUT_DIR -n NEOEPISCOPE_DATA_DIR -c HAPLOTYPE_DIR -d 72
python phasing_analysis.py -o OUTPUT_DIR -n NEOEPISCOPE_DATA_DIR -c HAPLOTYPE_DIR -d 94
The HAPLOTYPE_DIR is the directory containing the prepped haplotype files from HapCUT2/neoepiscope merge used to predict neoepitopes. For consistency with the script, the naming convention on these files should be PATIENT_ID.TUMOR_SAMPLE_ID.hapcut.out.prepped. The NEOEPISCOPE_DATA_DIR is the directory into which neoepiscope download saves it's data - make sure that you have run the downloader and selected 'yes' to download the hg19 GTF and bowtie index files.
To produce summary statistics and figures, we used the R script phasing_stats.R.
RNA-seq phasing evidence
For patients that had tumor RNA sequencing samples corresponding to tumor WES samples, we used STAR v2.6.1c to align reads to the hg19 genome.
First we indexed the genome:
STAR --runMode genomeGenerate --genomeDir STAR_INDEX_DIRECTORY --genomeFastaFiles REFERENCE_FASTA --sjdbGTFfile REFERENCE_GTF
Then we aligned reads:
STAR --runMode alignReads --outSAMattributes NH HI AS nM MD --outSAMstrandField intronMotif --outFileNamePrefix OUTPUT_DIRECTORY --genomeDir STAR_INDEX_DIRECTORY --readFilesCommand zcat --readFilesIn FASTQ1 FASTQ2
OUTPUT_DIRECTORY should be a directory named with the tumor RNA sample ID, as STAR does not specify output file names. We then sorted the resulting SAM files using samtools and converted to BAMs, which we indexed:
samtools sort -O BAM -o OUTPUT_BAM OUTPUT_DIRECTORY/Aligned.out.sam
samtools index OUTPUT_BAM
OUTPUT_BAM should be formatted as TUMOR_RNA_SAMPLE_ID.sorted.bam
We then used the script paired_read_rna_support.py to identify variant paired that were covered by the same RNA-seq read pair:
python paired_read_rna_support.py -p PATIENT_ID,TUMOR_WES_SAMPLE_ID,TUMOR_RNA_SAMPLE_ID -s STAR_OUTPUT_DIRECTORY -c HAPCUT_OUTPUT_DIRECTORY -o OUTPUT_DIRECTORY -d PICKLED_DICTIONARY
STAR_OUTPUT_DIRECTORY should be the directory containing your per-sample STAR output directories, HAPCUT_OUTPUT_DIRECTORY should be the directory containing your output from HapCUT2, OUTPUT_DIRECTORY is the directory to write output from this script, and PICKLED_DICTIONARY is created from running phasing_analysis.py with a distance of 72 bp - it should be in your output directory specified for that script, under the file name 'patient_variants_72.pickle'. This script will generate pickled dictionaries for each patient storing variant pairs with phasing supported by, not supported by, or novel in RNA-seq. We used counts from these to generate a tab separated text file for use in statistical analysis.
-
Wood MA, Nguyen A, Struck AJ, Ellrott K, Nellore A, Thompson RF. neoepiscope improves neoepitope prediction with multi-variant phasing. Bioinformatics. 2019;btz653.
-
Bassani-Sternberg M, Bräunlein E, Klar R, Engleitner T, Sinitcyn P, Audehm S, et al. Direct identification of clinically relevant neoepitopes presented on native human melanoma tissue by mass spectrometry. Nat Commun. 2016;7: 13404.
-
Amaria RN, Reddy SM, Tawbi HA, Davies MA, Ross MI, Glitza IC, et al. Neoadjuvant immune checkpoint blockade in high-risk resectable melanoma. Nat Med. 2018;24: 1649–1654.
-
Carreno BM, Magrini V, Becker-Hapak M, Kaabinejadian S, Hundal J, Petti AA, et al. A dendritic cell vaccine increases the breadth and diversity of melanoma neoantigen-specific T cells. Science. 2015;348: 803–808.
-
Gao J, Shi LZ, Zhao H, Chen J, Xiong L, He Q, et al. Loss of IFN-γ Pathway Genes in Tumor Cells as a Mechanism of Resistance to Anti-CTLA-4 Therapy. Cell. 2016;167: 397–404.e9.
-
Hugo W, Zaretsky JM, Sun L, Song C, Moreno BH, Hu-Lieskovan S, et al. Genomic and Transcriptomic Features of Response to Anti-PD-1 Therapy in Metastatic Melanoma. Cell. 2017;168: 542.
-
Roh W, Chen P-L, Reuben A, Spencer CN, Prieto PA, Miller JP, et al. Integrated molecular analysis of tumor biopsies on sequential CTLA-4 and PD-1 blockade reveals markers of response and resistance. Sci Transl Med. 2017;9.
-
Snyder A, Makarov V, Merghoub T, Yuan J, Zaretsky JM, Desrichard A, et al. Genetic basis for clinical response to CTLA-4 blockade in melanoma. N Engl J Med. 2014;371: 2189–2199.
-
Van Allen EM, Miao D, Schilling B, Shukla SA, Blank C, Zimmer L, et al. Genomic correlates of response to CTLA-4 blockade in metastatic melanoma. Science. 2015;350: 207–211.
-
Zaretsky JM, Garcia-Diaz A, Shin DS, Escuin-Ordinas H, Hugo W, Hu-Lieskovan S, et al. Mutations Associated with Acquired Resistance to PD-1 Blockade in Melanoma. N Engl J Med. 2016;375: 819–829.
-
Rizvi NA, Hellmann MD, Snyder A, Kvistborg P, Makarov V, Havel JJ, et al. Cancer immunology. Mutational landscape determines sensitivity to PD-1 blockade in non-small cell lung cancer. Science. 2015;348: 124–128.
-
Le DT, Durham JN, Smith KN, Wang H, Bartlett BR, Aulakh LK, et al. Mismatch repair deficiency predicts response of solid tumors to PD-1 blockade. Science. 2017;357: 409–413.