By proteomics approaches, the identification of HLA loaded peptides require multiple sequential steps because of the relative low abundance of HLA molecules (except in APCs) and a low abundance of loaded peptides; among the particular features (such as size, fixed positions of hydrophobic/hydrophilic moieties, PTMs,…) of the loaded peptides derived from protein degradation (proteosome,…) and assembled in HLA groove…. Hence, it seems that the separation and enrichment of HLA molecules is critical, and the selective elution of loaded peptides is also a key point. Thus, several methodological strategies have been performed and here it is discussed a few of them; despite of it is still an area in continuous change, progress and evolution ( Figure 2 ).

Most of the performed approaches are based on the specific and selective enrichment of HLA complexes by immuno-chromatography. Then, the HLA complex is immunopurified (IP) from the cell lysate (in presence of mild detergents), and further elution of peptides from the captured HLA complexes, which could be further analyzed, at high-resolution conditions, by liquid chromatography–mass spectrometry/mass spectrometry(LC-MS/MS). For this process, it is absolutely necessary to use optimal antibodies. Here, the most used are pan-HLA-I (Anti-Human HLA A, B, C: clone W6/32) and pan-HLA-II (Anti-Human HLA DR, DP, DQ: clone Tü39) (Bassani-Sternberg et al., 2010; Chong et al., 2018).

Regarding immunopurification, Chloe Chong et al. (2018) described a method based on sequential HLA purification by a chromatographic combination, starting by a pro-A beads (to capture endogenous antibodies), followed by an anti-HLA-I and/or anti-HLA-II antibodies coupled to pro-A beads to capture HLA class I and II, respectively. Then, HLA-I and HLA-II complexes are eluted and collected on a hydrophobic resin (C18) in order to selectively enrich loaded HLA peptides. The principal advantage of this strategy is that there are no intermediate steps, so it can continuously depleted endogenous antibodies and immunoaffinity purified class I and class II HLA complexes. Following this strategy, Chloe Chong et al. (2018) used B and T human cell lines and identified a total of 42,556 singular HLA class I peptides correlated to 8,975 proteins and 43,702 unrepeated HLA class II peptides from 4,501 proteins with a 1% false discovery rate (FDR). In both types of cell lines, the number of distinctive peptides changed from 3,293 to 13,696 for HLA I peptides and from 7,210 to 10,060 peptides for HLA II.

Another method is the direct immunoprecipitation using magnetic microspheres conjugated with anti-pan HLA I and II antibodies, respectively (Bassani-Sternberg et al., 2010; Chong et al., 2018). This procedure has the advantage of concentrating the cleft peptides of the HLA system thanks to the employment of super-paramagnetic microspheres conjugated to these highlighted antibodies. After separation of the peptides from the HLA molecules, the samples are eluted with an acid buffer and then the peptides are sequenced by mass spectrometry. If the whole exome or genome of the sample of interest is previously analysed to identify somatic mutations, then the results could be compared with the complete proteome (Chong et al., 2018; Kalaora and Samuels, 2019). In these studies, they compared several cancer cell lines with healthy ones, identified thousands of soluble HLA peptides, including some cancer specific peptides, shared among multiple several cell lines.

Recently, a novel procedure has been developed based on a mild acidic elution (MAE) of the HLA loaded peptides directly, in one single step, from the cells without any previous cell lysis and HLA selective enrichment. Here, thanks to the elution buffer thrown on the cells, the peptides carried by the cells of interest could be detected by further LC-MS/MS analysis. MAE strategy may be a cheap alternative to the other methods of immunoaffinity but is hindered by a large number of contaminating peptides not related with HLA molecules. By this strategy, Sturm T. et al., has identified the 50% common peptides between both approaches (MAE and immunoaffinity chromatography) as well as 22% of peptides identified only with MAE strategy (Sturm et al., 2020).

Having in mind that all human cells present HLA complexes and the critical role of HLA complexes in pathogen response and pathologies he precise and deep characterization of the immunopeptidome is highly important; hence, a synergically combination with multi-omics analysis (ie. metabolomics, genomics, proteomics, transcriptomics, epigenomics, among others) seems to be a powerful strategy for systematic determinations of immunopeptidomes. In addition to all of them, in silico prediction is a fundamental initial step to identify potential target neoantigens. For this purpose, it is crucial to carry out bioinformatic analysis that correlated the different databases and repositories of interest and integrate information from the immunopeptidome characterization ( Figure 3 ).

Currently, several open-access bioinformatics tools are available for the prediction and selection of neoantigens by personalized proteogenomic workflows. One of them is called ProGeo-Neo (https://github.com/kbvstmd/ProGeo-neo) that allows neoantigen prediction and selection based on a customized proteogenomic pipeline (Li et al., 2020). ProGeo-Neo is based on the integration of three dataset packages: i.-RNA-seq data analysis, which could generate variant peptides; ii-HLA alleles are inferred from RNA-seq data; then it is possible to work only on selected HLA alleles; iii.- Neoantigen prediction is based on genomic and proteomics information which allows the screening of new antigens by LC-MS/MS and a neoantigen filtering through RNA expression and T cell receptor recognition (epitope). This novel pipeline is already being used as in Xiaoxiu Tan’s study, where a platform was developed to facilitate the screening and confirmation of potential neoantigens in cancer immunotherapy (Tan et al., 2020).

In a similar manner, several databases have been recently created for identified peptides assembled in HLA complexes; mainly based on the employed methodology (commonly LC-MS/MS) for the identification of the peptides. Herein, it is briefly described some relevant ones with relevance to oncology and infections ( Table 1 ).

TRON Cell Line Portal or TCLP (http://celllines.tron-mainz.de/) (Scholtalbers et al., 2015) is a database that integrated the public RNA-Seq datasets of different exposed cell lines available in two repositories: The first set of data collected by Klijn et al. (2015) and the second one in The Cancer Cell Line Encyclopedia (CCLE). This database has been able to re-analyze accessible raw RNA-Seq datasets, determined the abundance and the type of HLA molecules as well as recognized virus and quantify the gene expression of 1,082 human cancer cell lines. Using all these available datasets of established HLA isotypes, cell lines- mutations and HLA prediction algorithms, Tron Cell Line Portal allows to predict the antigenic mutations in each analyzed human cell line. There are several studies in which they did typing of Human Leukocyte Antigens by High Throughput DNA and RNA Sequencing using TCLP, which includes an overview of approaches using high-throughput sequencing for HLA typing, as well as providing supplementary wet-lab protocols and in silico screening tools (Bukur, 2017; Boegel and Castle, 2019).

In this database, seq2HLA v2.2 is useful to identify the HLA isotypes (Boegel et al., 2014) which is able to calculate the four-digit HLA type from the RNA-Seq reads. It also generates two- and four-digit calls (Boegel et al., 2013; Boegel et al., 2014) with high precision. This public data includes HLA type data investigated by Adams et al. (2005), where sequence-based typing method (SBT) is used for HLA typing to determine the HLA class I and class II genotypes of the 60 cell lines used by National Cancer Institute (NCI-60). Using established HLA class I and II types in combination with systematic mutations will allow us to describe a register of possible neoepitopes candidates of HLA class I and II, respectively.

HLAthena (http://hlathena.tools/) This open-access portal is based on identification of HLA-I loaded peptides by exhaustive and systematic LC-MS/MS characterization. Here, more than 185,000 eluted peptides were analyzed from HLA A, B, C and G of 95 monoallelic human cell lines. The typical peptide motifs of each HLA allele was determined, as well as the unique and shared binding submotifs across alleles and the ones associated with a different peptide length. In addition, this database is developed by combining datasets with transcript abundance and the knowledge of peptide processing, which provides some prediction models of specific allele length for endogenous peptide presentation. These models predict HLA class I peptides compared with existing ligands tools and identify more than 75% of HLA I loaded peptides studied in 11 tumoral patients’ cells (melanoma, glioblastoma and clear cell renal cell carcinoma) with high accuracy. In summary, HLAthena database allows to systematically decipher the rules of presentation of endogenous antigens in tumoral cells (Abelin et al., 2017; Sarkizova et al., 2020). Other studies have used this database to predict and reduce the number of peptides to be studied in the face of the urgent need to develop a SARS-CoV-2 vaccine because the stimulation of an adequate immune response leading to protection is highly dependent on the presentation of epitopes to circulating T cells via the HLA complex. In this study 174 SARS-CoV-2 epitopes with high binding prediction scores were identified and validated to stably bind to 11 HLA allotypes (Prachar et al., 2020).

NetMHC 4.1 (http://www.cbs.dtu.dk/services/NetMHC/) is a method for predicting peptides bound to MHC class I using gap sequence alignment. This method is based on artificial neural network to align the amino acid sequence of peptides assembled in the HLA complexes, which allows insertion and deletion in the alignment. Alignment-based prediction methods including deletions and insertions show higher throughput than strategies trained on single-length peptides (Buus et al., 2003; Nielsen et al., 2003; Andreatta and Nielsen, 2016). Similarly, they exemplify how the position of the deletion can help explain the peptide-MHC binding pattern, such as when a long peptide protrudes from the HLA groove or protrudes at either end. And they also demonstrated that this method can predict the length distribution of different HLA molecules, and used this prediction algorithm to quantify the reduction in the experimental workload required to identify potential epitopes. There are several studies that use this database to carry out studies of very diverse pathologies (Conley et al., 2018; Khanna and Rana, 2019). As an example, one study uses this database to predict T-cell epitopes of Mycobacterium tuberculosis. As the identification of T-cell or B-cell epitopes on the target antigen is the main goal in epitope-based vaccine design, immunological diagnostic tests and antibody development, it is essential to provide a robust and reproducible system that can assist in the diagnosis of M. tuberculosis (Khanna and Rana, 2019).

SYFPEITHI (http://www.syfpeithi.de/) This approach contains a collection of MHC class I and II ligands and peptide motifs from humans as well as additional species (such as apes, cows, chickens, and mice), which is constantly being updated. You can search for HLA alleles and motifs, and you can find natural ligands, T cell epitopes, source of proteins and organisms and their references. It includes links to The European Molecular Biology Laboratory (EMBL) and PubMed databases. In addition, ligand prediction can be used for various HLA allele products. The content of the database is limited to the availability of published datasets; however, it is highly useful to correlate the prediction of T cell epitopes and HLA-loaded peptides (Rammensee et al., 1999).

The prediction based on published motifs (such as natural ligands and cluster sequencing) considers the amino acids at anchor positions. Calculation of a reliable score is for anchoring is done by giving some amino acids of determined peptides specific values ​​depending on whether they are anchors, auxiliary anchors or preferred residues. Ideal anchoring is 10 points, unconventional anchoring is 6 to 8 points, auxiliary anchoring is 4 to 6, and residues are preferably 1 to 4 points. Also, some amino acids which have a molecular weight that is considered to have a negative outcome on binding capacity are between −1 and −3 (Rammensee et al., 1999). This novel database has already been used in the characterization (previously discussed) of HLA ligandome in chronic lymphocytic leukemia, among others (Kowalewski et al., 2015).

IPD-IMGT/HLA (https://www.ebi.ac.uk/ipd/imgt/hla/) includes 25,000 allelic sequences of more than 40 genes, which are encoded by the MHC of the human genome. The IPD-IMGT/HLA is a stable, highly accessible and easy-to-use database which provides access to many alternative sequences of this genetic system to the medical and scientific communities, essential for, for example, successful transplant results. The challenge for this database is to continue providing a highly selected sequence variation database while keeping an increased amount of submissions and the complexity of the sequence (Robinson et al., 2020).

This database has multiple tools, either its own or that have been incorporated from data libraries to the existing tools available from the European Bioinformatics Institute (https://www.ebi.ac.uk/) (EMBL-EBI) (Valentin et al., 2010). Among them, you can find: i.-Sequence alignment, ii.-Allele query, iii.-sequence search tool (FASTA and BLAST) and iv.-cell query (original materials) (Labarga et al., 2007; Valentin et al., 2010; Robinson et al., 2014; Albrecht et al., 2017; Madeira et al., 2019).

The Immune Epitope Database (IEDB) (https://www.iedb.org/home_v3.php) is an open-access database supported by the National Institute of Allergy and Infectious Diseases (NIAID). It catalogues, in the context of infectious disease, allergy, autoimmunity and transplantation, different experimental studies on antibody and T cell epitopes examined in humans as well as non-human primates and several animal species (Vita et al., 2019). The IEDB also hosts tools to help in the prediction and analysis of epitopes. In one recent study, they use the IEDB database to design a model for vaccine design by prediction of B-epitopes in perturbations in the sequence of a multitude of peptides, in various source or host organisms (González-Díaz et al., 2014).

In addition to the information repositories about the peptides assembled in HLA molecules, tools for analyzing data extracted from LC-MS/MS characterization (including the different MS/MS instrumentation) and databases are increasingly common, including:

DeepRescore (https://github.com/bzhanglab/DeepRescore) is an immunopeptidomics data analysis device that supports deep learning-derived peptide characteristics to arrange peptide-spectrum matches (PSMs). It could take MS/MS raw data (in MGF format, for example) and recognize results from search engines as input. The last version supports four well-establish search engines like MS-GF+, Comet, X! Tandem and MaxQuant (Li et al., 2020). DeepRescore includes peptide characteristics derived from deep learning predictions (among which stand accurate retention time and MS/MS spectrum prediction) with previously used functions to reproduce peptide spectrum matching. Using two public immunopeptideomics data sets, it is demonstrated that compared with existing methods, the scoring performed by DeepRescore increases the accuracy and sensitivity of the recognition of MHC binding peptides and neoantigens. It also shows that, to a large extent, performance improvements are driven by features derived from deep learning. Thus, this post‐processing tool is freely available to the scientific community and can be used to identify sensitive and reproducible HLA binding peptides and neoantigens from immunopeptidomic datasets.

PEAKS X PRO (https://www.bioinfor.com/ Bioinformatics Solutions Inc., Waterloo, Ontario, Canada) is a commercially available software platform with several tools for the detection of peptide characteristics. Here, the three main characteristics are: DeepNovo is an extensive neural network model for de novo peptide sequencing. The DeepNovo design includes the latest developments in convolutional neural networks (CNN) and recurrent neural network (RNN) to learn the characteristics of peptides, as well as fragment ions and sequence patterns of tandem mass spectra. These systems are also integrated with local dynamic programming to solve the complex improvement tasks of de novo sequencing. DeepNovo has an accuracy growth at the amino acid level by 7.7% to 22.9% and an accuracy improvement at the peptide level by 38.1% to 64.0%. DeepNovo was used to automatically construct again the entire sequence of the mouse antibody light and heavy chain, without the need for others auxiliary databases to achieve 97.5% to 100% coverage and 97.2% to 99.5% accuracy. In addition, it can be retrained to modify any data source and supplies a whole end-to-end training and prediction solution for de novo sequencing problems (Tran et al., 2017). The second one is DeepIso, which combines the latest developments in CNN and RNN to estimate the peptide intensity and detect their features of different charge states. DeepIso consists of two different deep learning-based components, which can learn multiple levels of the high-dimensional data itself represented by multiple layers of neurons and can be adapted to new situations. The peptide characteristics list investigated with this model matches with 97.43% of high quality MS/MS identifications in a standard dataset (Zohora et al., 2019). The third one is DeepNovo-DIA, de novo peptide sequencing Independent data acquisition (DIA) method of mass spectrometry data. It also uses a neural network to capture cross m/z, retention time and intensity. Besides, DIA combined with peptide sequence pattern solves height problems multiple spectra, allowing us to identify novel peptides in human antigens and antibodies (Tran et al., 2019).

MaxQuant (https://www.maxquant.org/) is a quantitative proteomics software package developed for analyzing large mass-spectrometric data sets. It is the most used software with this type of raw data due to its versatility for handling and the interesting options with quantitative results. It is composed by a set of algorithms that extract information from raw MS data with high efficiency and sturdiness, capable of count elevated peptide identification rates, as well as high precision protein quantification for thousand proteins in complex proteomes (Cox and Mann, 2008). In recent years, the demand of MaxQuant has increased due to the great development of methodological strategies to identify peptides of interest in different pathologies (Schaab et al., 2012; Tyanova et al., 2015; Tyanova et al., 2016).