neoANT-HILL requires a variant list for potential neoantigen prediction. Our pipeline is able to handle a VCF file (single- or multi-sample) for the genome data or a tumor transcriptome sequence data (RNA-seq) in which somatic mutation will be called following GATK best practices [14-15] with Mutect2 [16] on tumor-only mode. However, the RNA-seq data must be previously aligned to the reference genome (BAM) by the user. The size of corresponding BAM files from the RNA-Seq can be a limiting factor in the analysis. Since neoANT-HILL is run locally, the user must guarantee that enough space and memory are available for a proper execution of the program. In the current implementation, neoANT-HILL supports VCF files generated using the human genome version GRCh37. The variants are properly annotated by snpEff [17] to identify non-synonymous mutations (missense, frameshift and inframe).
Once the VCF files have been annotated, the resulting altered amino acid sequences are inferred from the NCBI Reference Sequence database (RefSeq) [18]. For frameshift mutations, the altered amino acid sequence is inferred by translating the resulting cDNA sequence. Altered epitopes (neoepitopes) are translated into a 21-mer sequence where the altered residue is at the center. If the mutation is at the beginning or at the end of the transcript, the neoepitope sequence is built by taking the 20 following or preceding amino acids, respectively. The neoepitope sequence and its corresponding wild-type are stored in a FASTA file. Non-overlapping neoepitopes can be derived from frameshift mutations.
A list of HLA haplotypes is also required. If this data had not been provided by the user, neoANT-HILL includes the Optitype algorithm [19] to infers class-I HLA molecules from RNA-Seq. The subsequent step is the binding affinity prediction between the predicted neoepitopes and HLA molecules. This can be executed on single or multi-sample using parallelization with the custom configured parameters. The correspondent wild-type sequences are also submitted at this stage, which allows calculation of the fold change between wild-type and neoepitopes binding scores, known as differential agretopicity index (DAI) [as proposed by 20]. This additional neoantigen quality metric contributes to a better prediction of neoantigens that can elicit an antitumor response [21].
neoANT-HILL employs seven binding prediction algorithms from Immune Epitope Database (IEDB) [22], including NetMHC (v. 4.0) [23-24], NetMHCpan (v. 4.0) [25], NetMHCcons [26], NetMHCstabpan [27], PickPocket [28], SMM [29] and SMMPMBEC [30], and the MHCflurry algorithm [31] for HLA class I. The user is able to specify the neoepitope lengths to perform binding predictions. Each neoepitope sequence is parsed through a sliding window metric. Our pipeline also employs four IEDB-algorithms for HLA class II binding affinity prediction: NetMHCIIpan (v. 3.1) [32], NN-align [33], SMM- align [34], and Sturniolo [35].
Moreover, when the unmapped RNA-seq reads are available (fastq), neoANT-HILL can quantify the expression levels of genes carrying a potential neoantigen. Our pipeline uses the Kallisto algorithm [36] and the output is reported as transcripts per million (TPM). Potential neoantigens arising from genes showing an expression level under 1 TPM are excluded. In addition, neoANT-HILL also offers the possibility of estimating quantitatively, via deconvolution, the relative fractions of tumor-infiltrating immune cell types through the use of quanTIseq [37].
Our software was developed under a pre-built Docker image. The required dependencies are packed up, which simplify the installation process and avoid possible incompatibilities between versions. As previously described, several analyses are supported and each one relies on different tools. Several scripts were implemented on Python to complete automate the execution of these single tools and data integration.