NeoSplice is a novel computational method for splice variant neoantigen prediction based on 1) prediction of tumor-specific k-mers from RNA-seq data, 2) alignment of differentially expressed k-mers to the splice graph, and 3) inference of the variant transcript with MHC binding prediction. NeoSplice demonstrates high sensitivity and precision (>80% on average across all splice variant classes) through in silico simulated RNA-seq data. Additionally, predicted splice variant neoantigens in the K562.A2 cell line have been validated using mass spectrometry immunopeptidome analysis. NeoSplice provides a well-validated platform for prediction of TSA vaccine targets for future cancer antigen vaccine studies to evaluate the clinical efficacy of splice variant neoantigens.
The NeoSplice workflow is currently set up for prediction of MHC-I antigens, with future plans to additionally expand to MHC-II prediction.
NeoSplice is free for academic and non-profit use.
- multi-string BWT (MSBWT): https://github.com/holtjma/msbwt
- MSBWT-IS: https://github.com/holtjma/msbwt-is
- NetMHCpan 4.0
- Python 2.7
- networkx 1.11
- pyahocorasick 1.4.0
- bcbio-gff 0.6.4
- pyfaidx 0.5.3.1
- pysam 0.14.1
- biopython 1.70
- scipy 1.2.0
Using a Python 2.7 VirtualEnv:
git clone https://github.com/Benjamin-Vincent-Lab/NeoSplice.git cd NeoSplice virtualenv . source bin/activate pip install -r requirements.txt
Using Singularity with Docker image:
singularity pull docker://benjaminvincentlab/neosplice:0.0.2 singularity shell -B </path/with/input_bams_gffs_fa_etc> neosplice_0.0.2.sif
Multiple steps are needed to identify a novel splice that occurs specifically in tumor cell transcripts whose translation will result in a neopeptide that can be targeted by T cells. Broadly, these steps are:
- Step 1: Tumor-specific k-mer generation. Using one RNA-seq dataset T of tumor cells and one RNA-seq dataset N of normal cells, tumor-specific k-mer sequences present abundantly in the transcriptome of the tumor cell, but not/rarely expressed in the normal cells are identified.
- Step 2: Prediction of splice variant transcripts. The splice graph G from the tumor cell RNA-seq data is built, and tumor specific k-mers from above are mapped to novel splice variant transcripts. Gencode annotations are used to determine whether the novel splice lies within a protein coding region and infer the reading frame of the transcript.
- Step 3: Prediction of splice variant neoantigens. Novel splice junctions contained within each splice variant transcript are translated in the inferred open reading frame. MHC binding affinity prediction is performed on translated peptide sequences to determine which novel regions may yield a neopeptide.
Functionally, these above steps are accomplished by individual Python2 scripts, alongside the prior listed dependencies. This workflow is summarized in the below figure:
This workflow is summarized step-by-step below. Additionally, an example Nextflow script is provided in the ./Nextflow_example directory of the GitHub repo, which provides the entire workflow as an .nf script.
The following input files will be referenced in the below workflow steps:
- tumor.bam: Aligned RNA-seq file for tumor sample of interest using a splice-aware aligner. Currently, we recommend the STAR aligner: https://github.com/alexdobin/STAR
- normal.bam: Aligned RNA-seq file for matched-normal sample of interest using a splice-aware aligner. Currently, we recommend the STAR aligner: https://github.com/alexdobin/STAR
- reference.fa: Reference genome fasta file, preferably the same file used to generate the genome index for STAR.
- reference.gff: Reference gff3 file, preferably the same build as the .gtf file that was used to generate the genome index for STAR.
- ./Reference_peptidome: A GRCh38 reference peptidome of 8-11mer peptides is contained within the repository at
./Reference_peptidome. If necessary, you can use the script./NeoSplice/generate_reference_peptidome.pyto generate a different reference build using matching fasta and gff files. Prior to use (step 9), unzip the files contained in this directory using the following command:
gunzip ./Reference_peptidome/*
This step builds the splice graph for the tumor, with augmented_splice_graph.py run for each individual chromosome of interest. The output for each instance (i.e. chromosome) is a .json file. There are several arguments included for this step: p-error, cutoff, min-coverage, and min-variants. While we cannot provide optimal argument recommendations for every sample, below are the values used for simulated read data benchmarking and mass spectrometry validated K562.A2 cell line splice variant neoantigens. Below is an example for chromosome 1:
mkdir ./tumor1_splice_graph
python /NeoSplice/augmented_splice_graph.py build \
--bam ./path/to/tumor.bam \
--seq chr1 \
--genome ./path/to/reference.fa \
--min-variants 10 \
--cutoff 0.000005 \
--gff ./path/to/reference.gff \
--out ./tumor_splice_graph
This step is a simple script to back-convert the STAR-aligned tumor.bam and normal.bam files back into fasta format:
python /NeoSplice/convert_bam_to_fasta.py \
--bam_file tumor.bam \
--R1_out tumor_R1.fasta \
--R2_out tumor_R2.fasta
python /NeoSplice/convert_bam_to_fasta.py \
-bam_file normal.bam \
--R1_out normal_R1.fasta \
--R2_out normal_R2.fasta
This step uses the MSBWT-IS tool developed by Holt and colleagues (https://github.com/holtjma/msbwt-is), followed by a bash script to convert the output format for downstream compatibility:
mkdir ./tumor_bwt/ mkdir ./normal_bwt/ mkdir ./tumor_bwt_temp/ mkdir ./normal_bwt_temp/ ./msbwt-is/msbwtis tumor_bwt_temp/ tumor_R1.fasta tumor_R2.fasta ./msbwt-is/msbwtis normal_bwt_temp/ normal_R1.fasta normal_R2.fasta bash ./NeoSplice/convert_BWT_format.bash ./tumor_bwt_temp ./tumor_bwt bash ./NeoSplice/convert_BWT_format.bash ./normal_bwt_temp ./normal_bwt
This step searches for the maximum read length contained within either the tumor or matched-normal files, returning an output value for use in step 5. If you know this value already, this step can be skipped:
python /NeoSplice/get_max_kmer_length.py \
--tumor_bam tumor.bam \
--normal_bam normal.bam
This step uses the MSBWTs generated in step 3 and searches for differentially expressed Kmers between tumor and matched-normal samples. There are two argument variables that can be adjusted here -- Tmin (minimum expression of a given Kmer in the tumor) and Nmax (maximum expression of a given Kmer in the normal). For a Kmer to be considered differentially expressed, it must be > Tmin AND < Nmax. Typically, you may consider setting Tmin to 20-35 and Nmax to 1-4. The max_length argument should be set to the value obtained from step 4, or the maximum read length of the input files.
mkdir .tumor_kmers
python ./NeoSplice/Kmer_search_bwt.py \
--tumor_bwt ./tumor_bwt/ \
--normal_bwt ./normal_bwt/ \
--processors 1 \
--max_length $read_length \
--tumor_threshold 20 \
--normal_threshold 4 \
--outdir ./tumor_kmers/
cat ./tumor_kmers/Tumor_kmers_* > ./tumor_kmers/merged_Tumor_kmers.txt
This step uses an Aho–Corasick algorithm (pyahocorasick 1.4.0) to search for the reads that contain tumor specific Kmers in the tumor RNA-seq BAM file. This method runs in time linear in the size of the BAM file. For each occurrence, the Kmer-containing portion of the read along with corresponding quality scores and Cigar strings is written to a new BAM file. This output BAM is then sorted and indexed using Samtools.
python ./NeoSplice/search_bam.py \
--Kmer_file ./tumor_kmers/merged_Tumor_kmers.txt \
--input_bam_file tumor.bam \
--out_bam_file tumor_Kmer.bam
samtools sort -m 15G -o tumor_Kmer_sorted.bam tumor_Kmer.bam
samtools index tumor_Kmer_sorted.bam
This step collects a list of all splice junctions from the tumor and normal BAM files, storing these in a text file for downstream use.
python /NeoSplice/get_splice_junctions.py \
--input_bam tumor.bam \
--out_file tumor_junctions.txt
python /NeoSplice/get_splice_junctions.py \
--input_bam normal.bam \
--out_file normal_junctions.txt
In this step, each splice variant transcript sequence is identified by depth-first search. This is then concatenated with the tumor specific Kmer sequence and translated into 8-11mer peptides for MHC-I neoantigen prediction. Binding affinity to MHC molecules expressed by the tumor for in-silico generated peptides is predicted using NetMHCpan-4.0. Arguments to consider in this step include HLA_I (provide list of NetMHCpan-compatible alleles for antigen prediction), as well as transcript_min_coverage (the minimum Kmer coverage necessary for a transcript to be considered). This command is run for each chromosome of interest, with an example for chromsome 1 shown below:
python /NeoSplice/kmer_graph_inference.py \
--sample tumor \
--chromosome chr1 \
--bam_file tumor.bam \
--gff_file reference.gff \
--genome_fasta reference.fasta \
--kmer_bam tumor_Kmer_sorted.bam \
--splice_graph ./tumor_splice_graph \
--tumor_junction_file tumor_junctions.txt \
--normal_junction_file normal_junctions.txt \
--transcript_min_coverage 15 \
--HLA_I ${HLA_i} \
--netMHCpan_path ./netMHCpan-4.0-docker/netMHCpan \
--outdir ./tumor_output_dir
In this final step, predicted splice variant peptides from above are filtered against the reference peptidome, filtered to peptides with predicted binding affinity >500nM by NetMHCpan-4.0, and summarized into a single output file. The data_dir argument should point to the working directory, one level above the outdir argument from step 8 (kmer_graph_inference.py). The output from this step provides a summarized text file containing all predicted splice variant neoantigens.
python /NeoSplice/SV_summarization.py \
--ref_dir ./Reference_peptidome \
--data_dir .