Comparison of computational tools for neoantigen identification and immunogenicity prediction
| Tool | Step | Data input | Algorithm type | Key strength | Limitations | References |
|---|---|---|---|---|---|---|
| INTEGRATE-neo | Mutation calling | Whole exome/Genome sequencing | Graph-based detection | High sensitivity for detecting insertions and complex mutations | Limited to specific mutation types; may miss small variants | [218–220] |
| Polysolver | HLA typing | Whole exome sequencing | Bayesian inference | Highly accurate even with low-coverage data | Computationally intensive, requires high processing power | [201–203] |
| NetMHCpan | HLA binding prediction | Peptide sequences | Neural network | Broad coverage of MHC alleles across diverse populations | Sequence length constraints may not predict all peptides | [221–224] |
| NetMHCIIpan | HLA binding prediction | Peptide sequences | Neural network | High sensitivity for MHC class II binding predictions | Limited coverage of MHC class II alleles, not exhaustive | [225, 226] |
| NetCTL | T cell recognition | Peptide-HLA binding data | Machine learning | Specific scoring for CD8+ T cell recognition | Does not model TCR structure or interactions well | [227] |
| MHCflurry | HLA binding prediction | Peptide sequences | Machine learning/Deep learning | Accurate binding affinity predictions for class I peptides | Limited prediction capability for rare MHC alleles | [228, 229] |
| VAXign | Epitope validation | Peptide sequences | Statistical modeling | High throughput capability for large peptide datasets | Requires large amounts of experimental validation data | [230–232] |
| Immune Epitope Database (IEDB) | Epitope validation | Peptide sequences, HLA typing | Database query | Extensive experimental data support, widely recognized database | Limited by available datasets, may lack specificity in certain cases | [233–235] |
| NeoepitopePred | T cell recognition | Peptide-HLA binding data | Hybrid model (SVM, neural networks) | Comprehensive T-cell response prediction model | Requires large training datasets, may overestimate some responses | [172, 236, 237] |
| MHC-I Binding Prediction Tool | HLA binding prediction | Peptide sequences | Weighted scoring system | Effective for a wide range of peptide sequences | Limited by scoring system, may not capture all binding interactions | [238–240] |
HLA: human leukocyte antigen; TCR: T-cell receptor; SVM: support vector machines; MHC: major histocompatibility complex
