Bio Immunoinformatics Mhc Class Ii Prediction

Version Compatibility

Reference examples tested with: NetMHCIIpan 4.3+, MixMHC2pred 2.0+, pandas 2.2+

Before using code patterns, verify installed versions match. If versions differ:

• Python: pip show <package> then help(module.function) to check signatures
• CLI: <tool> --version then <tool> --help to confirm flags

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

Notes specific to this skill: NetMHCIIpan and MixMHC2pred are standalone academic binaries (not pip-installable); the IEDB MHC-II REST API wraps NetMHCIIpan if a local install is unavailable. Allele nomenclature differs sharply between tools and between isotypes (DR single-chain vs DQ/DP heterodimer) — confirm the exact string format against the installed tool before scoring. Class II %Rank thresholds (1%/5%) are LOOSER than class I (0.5%/2.0%); do not copy class I cutoffs.

MHC Class II Prediction

"Predict which long peptides bind/are presented by HLA class II" -> Score peptide-HLA class II (DR/DQ/DP) presentation to nominate candidate CD4 T-cell epitopes, inferring the 9-mer binding core within each long peptide.

• CLI: NetMHCIIpan (field default; EL score by default, -BA adds affinity; pan-DR/DQ/DP)
• CLI: MixMHC2pred (MS-deconvolution motifs; models the reverse DP binding mode)

The Single Most Important Modern Insight -- class II is basically broken, and that must be stated plainly

For class I, modern pan-allele EL predictors recover most true ligands at high precision and the field has hit diminishing returns. The same architectures, on the same conceptual pipeline, produce dramatically weaker class II predictions. The honest one-line summary to give a collaborator is: "trust a class I strong-binder call; treat a class II call as a ranked hypothesis, not a fact." Four compounding reasons, not one, cause this. The groove is open at both ends, so a 12-25mer can sit in multiple registers and the model must infer which latent 9-residue core is the true binding frame — an error-prone latent-variable problem class I (closed groove, defined termini) never faces. The training data are smaller and noisier: in-vitro class II binding assays are notoriously irreproducible, and class II immunopeptidomics yields fewer, longer, more heterogeneous peptides. The three isotypes are unequally tractable — historically DR >> DP > DQ, because DR was studied first and most while DQ was data-starved (NetMHCIIpan-4.3's headline 2023 contribution was finally closing this gap with tailored data acquisition, a sign of how recent and data-driven the fix is). And DP/DQ are obligate alpha/beta heterodimers whose chains are independently polymorphic, so the effective number of distinct molecules is the combinatorial product of alpha and beta alleles.

The DQ/DP heterodimer pairing trap

A donor's DQA1 and DQB1 alleles pair both in cis (same haplotype) and in trans (across haplotypes), so a heterozygous individual can express up to four DQ heterodimers — and some trans-pairs are non-functional or rare. Mechanically feeding all DQA1 x DQB1 combinations to NetMHCIIpan generates molecules that do not biologically exist; taking only cis pairs may miss real trans-dimers. There is no fully automated, universally agreed resolution. The expert move is to be explicit about the pairing assumption, prefer documented haplotype pairings, and flag DQ (and to a lesser extent DP) epitope calls as lower-confidence than DR. DR is single-chain (the alpha is effectively invariant), so it carries none of this combinatorial burden and is the most trustworthy isotype.

Tool Taxonomy (Class II)

Decision Tree by Scenario

Running the Predictions (the fragile commands - run as written)

NetMHCIIpan reads a peptide list or FASTA and scores against one or more alleles. EL %Rank is the default output; -BA adds an affinity prediction. The model reports the inferred 9-mer core and its offset.

# DR (single-chain: beta allele names the molecule)
netMHCIIpan -f peptides.txt -inptype 1 -a DRB1_0101 -BA -xls -xlsfile out.tsv

# DQ heterodimer (BOTH chains, hyphen-joined) and DP
netMHCIIpan -f antigen.fasta -a HLA-DQA10501-DQB10201,HLA-DPA10103-DPB10401 -length 15

Key flags: -a allele(s, comma-separated), -f input, -inptype (0=FASTA, 1=peptide list), -length peptide length(s) to consider, -BA add affinity, -xls/-xlsfile tab output, -list dump supported alleles.

MixMHC2pred uses chain-underscore allele names with a DOUBLE underscore between heterodimer chains, and alleles are space-separated:

MixMHC2pred -i peptides.txt -o out.txt -a DRB1_15_01 DRB5_01_01 DPA1_02_01__DPB1_01_01

Per-Method Failure Modes