NeoFuse is a user-friendly pipeline for the prediction of fusion neoantigens from tumor RNA-seq data.
NeoFuse takes single-sample FASTQ files of RNA-seq reads (single- or paired-end) as input and predicts putative fusion neoantigens through five main analytical modules based on state-of-the-art computational tools:
- Genotyping of class-I Human Leukocyte Antigen (HLA) genes at 4-digit resolution using OptiType (Szolek et al., 2014).
- Prediction of fusion peptides using Arriba (https://github.com/suhrig/arriba), together with confidence scores reflecting the likelihood that a fusion is caused by a tumor-specific genomic rearrangement and is not due to technical artifacts.
- Prediction of the binding affinity of fusion peptides to HLA types, quantified as half maximal inhibitory concentration (IC50) and percentile rank, using MHCflurry (O’Donnell et al., 2018) or netMHCpan (Jurtz et al., 2017).
- Quantification of gene expression levels, as transcripts per million (TPM), using STAR (Dobin et al., 2013) and featureCounts (Liao et al., 2014).
- Neoantigen prioritization based on IC50 binding affinity and confidence score, and annotation of each neoantigen with: IC50, percentile rank, confidence score, binding HLA type, expression of the fusion and HLA genes in TPM, and information about the presence of a premature stop codon that might cause nonsense mediated decay of the fusion transcript.
We advise using paired-end data to increase sensitivity and accuracy of gene fusion detection.
- At least 30 GB of RAM
- Docker (version 19.03 or later) or
- Singularity (version 3.0 or later)
NeoFuse can be installed through the following four steps.
Instructions for Docker installation
Instructions for Singularity installation
The script is freely available here.
Unzip the archive and add it to PATH:
$ export PATH=$PATH:~/path/to/NeoFuse
The NeoFuse image can be automatically generated using the NeoFuse script:
Docker:
$ NeoFuse -B --docker
Singularity:
$ NeoFuse -B --singularity
The NeoFuse script can be also used to generate the genomes and indexes required by the analysis:
Docker:
$ Neofuse -R -o </path/to/output_folder> -n [cores] -V [genome version] --docker
Singularity:
$ Neofuse -R -o </path/to/output_folder> -n [cores] -V [genome version] --singularity
<Arguments>
-o: Output directory
[Options]
-n: Number of cores (default: 1)
-V: Genome version, either “GRCh37” and “GRCh38” (default: GRCh38)
Note: this process may take more than 1 hour, depending on the internet connection and the processing power.
Notes
-
On Mac OS X, you need to have Docker running to execute NeoFuse.
-
On Mac OS X, you might need to increase CPUs, Memory and Swap in Docker settings (Settings > Preferences > Advanced).
NeoFuse can process single samples with the following command:
$ NeoFuse <arguments> [options] --singularity (or --docker)
<Arguments>
-1: Path to read 1 FASTQ file (mandatory)
-2: Path to read 2 FASTQ fie (optional for single-end reads)
-s: Path to STAR index directory (mandatory)
-g: Path to reference genome FASTA file (mandatory)
-a: Path to annotation GTF file (mandatory)
Note: All input files passed as arguments must be unzipped.
[Options]
-d: Run ID (the name of the output files)
-m: Minimum peptide length (values: 8, 9, 10, or 11; default: 8)
-M: Maximum peptide length (values: 8, 9, 10, or 11; default: 8) *
-n: Number of cores (default: 1)
-t: IC50 binding affinity threshold (default: 500)
-T: Percentile rank threshold (default: Inf)
-c: Mimimum confidence score (values: H, M, or L; default: L) **
--singularity: NeoFuse will use the Singularity image
--docker: NeoFuse will use the Docker image
* NeoFuse will compute the binding affinity for all the possible lengths of peptides between the minimum and maximum input. For example if a user specifies '-m 8' and '-M 11', NeoFuse will compute the binding affinity for all peptides of length 8, 9, 10, and 11. To consider just one specific length, use only the '-m' argument.
** The mimimum Arriba confidence score can be set to: H (to return only high confidence fusions), M (for high and medium confidence fusions), or L (for high, medium, and low confidence fusions).
For multiple-sample analysis, a TSV input file reporting the sample identifiers and path to input files has to be prepared. Format:
Paired-end reads:
#ID Read1 Read2
Sample1 /path/to/Sample1_read_1.fastq /path/to/Sample1_read_2.fastq
Sample2 /path/to/Sample2_read_1.fastq /path/to/Sample2_read_2.fastq
Single-end reads:
#ID Read1
Sample1 /path/to/Sample1_read_1.fastq
Sample2 /path/to/Sample2_read_1.fastq
Notes: The first line of the TSV should start with an hashtag. There should always be one blank row at the end of the TSV file.
Once the TSV file is created, the samples can be analyzed with the following command:
$ NeoFuse <arguments> [options] --singularity (or --docker)
<Arguments>
-i: Path to the input TSV file (mandatory)
-s: Path to STAR index directory (mandatory)
-g: Path to reference genome FASTA file (mandatory)
-a: Path to annotation GTF file (mandatory)
Note: All input files passed as arguments must be unzipped.
[Options]
-m: Minimum peptide length (values: 8, 9, 10, or 11; default: 8)
-M: Maximum peptide length (values: 8, 9, 10, or 11; default: 8) *
-n: Number of cores (default: 1)
-t: IC50 binding affinity threshold (default: 500)
-T: Percentile rank threshold (default: Inf)
-c: Mimimum confidence score (values: H, M, or L; default: L) **
--singularity: NeoFuse will use the Singularity image
--docker: NeoFuse will use the Docker image
* NeoFuse will compute the binding affinity for all the possible lengths of peptides between the minimum and maximum input. For example if a user specifies '-m 8' and '-M 11', NeoFuse will comppute the binding affinity for all peptides of length 8, 9, 10, and 11. To consider just one specific length, use only the '-m' argument.
** The mimimum Arriba confidence score can be set to: H (to return only high confidence fusions), M (for high and medium confidence fusions), or L (for high, medium, and low confidence fusions).
Due to license compatability issues, netMHCpan is fully integrated but not distributed as part of NeoFuse.
If there is an existing local installation of netMHCpan, peptide-HLA binding affinity (IC50 and rank) can be predicted with netMHCpan instead of MHCflurry using the following command:
$ NeoFuse <arguments> [options] -N [/path/to/netMHCpan_direcotry] --singularity (or --docker)
NeoFuse will create an output directory with the following structure:
/NeoFuse/output/directory/
├── Sample1
│ ├── Arriba
│ ├── LOGS
│ ├── NeoFuse
│ ├── OptiType
│ └── TPM
├── Sample2
│ ├── Arriba
│ ├── LOGS
│ ├── NeoFuse
│ ├── OptiType
│ └── TPM
…
└── SampleN
├── Arriba
├── LOGS
├── NeoFuse
├── OptiType
└── TPM
Sample.fusions.tsv file contains a list of gene fusions sorted from highest to lowest confidence.
Sample.fusions.discarded.tsv contains all events that Arriba classified as artifacts or that are also observed in healthy tissues.
/Arriba
├── Sample1.fusions.discarded.tsv
└── Sample1.fusions.tsv
The standard output (sdout and stderr) for every tool used in the run is stored in the LOGS directory. File names may differ depending on the tools, peptide length, etc.
/LOGS
├── Sample1_10_MHCFlurry.log
├── Sample1_11_MHCFlurry.log
├── Sample1_8_MHCFlurry.log
├── Sample1_9_MHCFlurry.log
├── Sample1.arriba.err
├── Sample1.arriba.log
├── Sample1.cleave_peptides.log
├── Sample1.counts_to_tpm.log
├── Sample1.featureCounts.log
├── Sample1.final.log
├── Sample1.Log.final.out
├── Sample1.Log.out
├── Sample1.Log.std.out
├── Sample1.optitype.log
├── Sample1.razer1.log
├── Sample1.razer2.log
├── Sample1.STAR.err
├── Sample1.STAR.log
└── Sample1.association.log
HLA_Optitype.txt contains the HLA types predicted by OptiType
/OptiType
└── Sample1_HLA_Optitype.txt
Contains all TPM expression values for all the genes
/TPM
└── Sample1.tpm.txt
Contains the final output of the pipeline, which consists of three files:
/NeoFuse
├── Sample1_filtered.tsv
├── Sample1_unfiltered.tsv
└── Sample1_unsupported.txt
Sample_unsupported.txt contains the HLA types predicted by OptiType that are not supported by MHCflurry. Note: if netMHCpan is used instead of MHCfurry, this file is not generated.
Sample_unfiltered.tsv contains all the predicted fusion peptides and their annotations.
Sample_filtered.tsv contains a list of putative fusion neoantigens (selected considering the user-defined IC50/rank and confidence score thresholds). This file reports for each putative neoantigen: confidence score, binding HLA type, expression of the fusion and HLA genes in TPM, and information about the presence of a premature stop codon that might cause nonsense mediated decay of the fusion transcript. Example format:
Fusion Gene1 Gene2 HLA_Type Fusion_Peptide IC50 Rank Event_Type Stop_Codon Confidence Gene1_TPM Gene2_TPM Avg_TPM HLA_TPM
BCAS4-BCAS3 BCAS4 BCAS3 HLA-A*02:01 FLTPDPGAEV 51.07703105715209 0.3441249999999996 Fusion-out-of-frame yes high 29.07 219.64 124.35 22.40
RSBN1-AP4B1-AS1 RSBN1 AP4B1-AS1 HLA-A*02:01 STLEAPLAPA 372.5722360454703 1.3481250000000005 Fusion-out-of-frame no high 15.99 5.10 10.54 22.40
PPM1D-USP32 PPM1D USP32 HLA-B*44:02 LEMGSPYVAQ 361.06034776084135 0.7239999999999998 Fusion-out-of-frame no high 129.58 178.94 154.26 28.59
TXLNG-SYAP1 TXLNG SYAP1 HLA-A*02:01 KTIYLTLHLL 281.7035889624228 1.1387500000000002 Fusion-out-of-frame yes low 21.70 46.29 33.99 22.40
NAV1-GPR37L1 NAV1 GPR37L1 HLA-A*02:01 RLSASVPWAL 18.46819203380198 0.11887500000000008 Fusion-out-of-frame no low 0.89 2.89 1.89 22.40
NAV1-GPR37L1 NAV1 GPR37L1 HLA-A*02:01 RLSASVPWAL 18.46819203380198 0.11887500000000008 Fusion-out-of-frame no low 0.89 2.89 1.89 22.40
Dobin,A. et al. (2013) STAR: ultrafast universal RNA-seq aligner. Bioinformatics, 29, 15–21.
Jurtz, V. et al. (2017) NetMHCpan-4.0: Improved Peptide–MHC Class I Interaction Predictions Integrating Eluted Ligand and Peptide Binding Affinity Data. J. Immunol., 199, 3360-3368.
Liao,Y. et al. (2014) featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics, 30, 923–930.
O’Donnell,T.J. et al. (2018) MHCflurry: Open-Source Class I MHC Binding Affinity Prediction. Cell Syst, 7, 129–132.e4.
Szolek,A. et al. (2014) OptiType: precision HLA typing from next-generation sequencing data. Bioinformatics, 30, 3310–3316.