Predictor matrix¶

Reference for supported predictors, Python classes, command-line names, inputs, and installation routes. For model selection, see choosing a predictor.

Kinds are defined in prediction kinds. A predictor emits only the kinds that apply (for example NetMHCpan41_EL emits presentation only).
MHC class is the class the model scores. None means the prediction is MHC-independent and Prediction.allele is empty.
Default lengths apply to predict_proteins() and the command line when you pass no lengths; they are narrower than what the model supports. See peptide lengths.
License tiers are defined in licensing; mhctools fetch behavior is in getting models.

MHC binding and presentation¶

See the mhc binding and presentation guide for examples and model notes.

Predictor	MHC class	Input
NetMHCpan 4.1 / 4.2	I	peptides + alleles
NetMHCpan 4.0	I	peptides + alleles
NetMHCpan 3.0 / 2.8	I	peptides + alleles
NetMHC 3.4 / 4.0	I	peptides + alleles
NetMHCcons	I	peptides + alleles
NetMHCIIpan 4.3 / 4.1 / 4.0 / 3.x	II	peptides + alleles
NetMHCstabpan	I	peptides + alleles
MHCflurry	I	peptides + alleles (flanks optional)
BigMHC	I	peptides + alleles
CapHLA	I and II	peptides + alleles
MixMHCpred	I	peptides + alleles
MixMHC2pred	II	peptides + alleles
SMM / SMM-PMBEC	I	peptides + alleles
Historical IEDB names	I and II	peptides + alleles
RandomBindingPredictor	I	peptides + alleles

NetMHCpan 4.1 / 4.2¶

Python classes: NetMHCpan, NetMHCpan41, NetMHCpan41_BA, NetMHCpan41_EL, NetMHCpan42, NetMHCpan42_BA, NetMHCpan42_EL
CLI names: netmhcpan, netmhcpan41, netmhcpan41-ba, netmhcpan41-el, netmhcpan42, netmhcpan42-ba, netmhcpan42-el
Prediction kinds: pMHC_affinity, pMHC_presentation
Default scanning lengths: 9
Installation: your licensed DTU install
License: dtu

NetMHCpan 4.0¶

Python classes: NetMHCpan4, NetMHCpan4_BA, NetMHCpan4_EL
CLI names: netmhcpan4, netmhcpan4-ba, netmhcpan4-el
Prediction kinds: pMHC_affinity, pMHC_presentation
Default scanning lengths: 9
Installation: your licensed DTU install
License: dtu

NetMHCpan 3.0 / 2.8¶

Python classes: NetMHCpan3, NetMHCpan28
CLI names: netmhcpan3, netmhcpan28
Prediction kinds: pMHC_affinity
Default scanning lengths: 9
Installation: your licensed DTU install
License: dtu

NetMHC 3.4 / 4.0¶

Python classes: NetMHC, NetMHC3, NetMHC4
CLI names: netmhc, netmhc3, netmhc4
Prediction kinds: pMHC_affinity
Default scanning lengths: 9
Installation: your licensed DTU install
License: dtu

NetMHCcons¶

Python classes: NetMHCcons
CLI names: netmhccons
Prediction kinds: pMHC_affinity
Default scanning lengths: 9
Installation: your licensed DTU install
License: dtu

NetMHCIIpan 4.3 / 4.1 / 4.0 / 3.x¶

Python classes: NetMHCIIpan, NetMHCIIpan3, NetMHCIIpan4, NetMHCIIpan4_BA, NetMHCIIpan4_EL, NetMHCIIpan43, NetMHCIIpan43_BA, NetMHCIIpan43_EL
CLI names: netmhciipan, netmhciipan3, netmhciipan4, netmhciipan4-ba, netmhciipan4-el, netmhciipan43, netmhciipan43-ba, netmhciipan43-el
Prediction kinds: pMHC_affinity, pMHC_presentation
Default scanning lengths: 15-20
Installation: your licensed DTU install
License: dtu

NetMHCstabpan¶

Python classes: NetMHCstabpan
CLI names: netmhcstabpan
Prediction kinds: pMHC_stability
Default scanning lengths: (none; pass lengths)
Installation: your licensed DTU install
License: dtu

MHCflurry¶

Python classes: MHCflurry, MHCflurry_Affinity
CLI names: mhcflurry, mhcflurry-affinity
Prediction kinds: pMHC_affinity, pMHC_presentation, antigen_processing
Default scanning lengths: 9
Installation: mhctools fetch mhcflurry
License: open

BigMHC¶

Python classes: BigMHC, BigMHC_EL, BigMHC_IM
CLI names: bigmhc, bigmhc-el, bigmhc-im
Prediction kinds: pMHC_presentation, immunogenicity
Default scanning lengths: (set per call)
Installation: mhctools fetch bigmhc --accept-license + PyTorch
License: academic

CapHLA¶

Python classes: CapHLA, CapHLA_BA, CapHLA_EL
CLI names: caphla, caphla-ba, caphla-el
Prediction kinds: pMHC_affinity, pMHC_presentation
Default scanning lengths: (7-25 supported)
Installation: pip install "mhctools[caphla]" + mhctools fetch caphla
License: open

MixMHCpred¶

Python classes: MixMHCpred
CLI names: mixmhcpred
Prediction kinds: pMHC_presentation
Default scanning lengths: 9 (8-14 supported)
Installation: upstream release + MIXMHCPRED_PATH
License: academic

MixMHC2pred¶

Python classes: MixMHC2pred
CLI names: mixmhc2pred
Prediction kinds: pMHC_presentation
Default scanning lengths: 15
Installation: upstream release (with PWMdef/)
License: academic

SMM / SMM-PMBEC¶

Python classes: SMM, SMMPMBEC
CLI names: smm, smm-pmbec
Prediction kinds: pMHC_affinity
Default scanning lengths: 9
Installation: python scripts/setup_test_backends.py smm --accept-license or IEDB_MHCI_EXECUTABLE
License: open

Historical IEDB names¶

Python classes: IedbNetMHCpan, IedbNetMHCcons, IedbNetMHCIIpan, IedbSMM, IedbSMM_PMBEC
CLI names: netmhcpan-iedb, netmhccons-iedb, netmhciipan-iedb, smm-iedb, smm-pmbec-iedb
Prediction kinds: pMHC_affinity
Default scanning lengths: 8-11 (class I), 15-20 (II)
Installation: the local predictor each one maps to
License: dtu

RandomBindingPredictor¶

Python classes: RandomBindingPredictor
CLI names: random
Prediction kinds: pMHC_affinity
Default scanning lengths: 9
Installation: built in
License: builtin

Antigen processing¶

See the antigen processing guide for examples and model notes.

Predictor	MHC class	Input
Pepsickle	none	peptides (flanks recommended)
NetChop	none	peptides (flanks recommended)
NetCleave	I or II	peptides + C-terminal flank (>= 3 residues)
DeepTAP	none	peptides only
ERAMER	none (class I context)	peptides only (9-16mer precursors)

Pepsickle¶

Python classes: Pepsickle
CLI names: pepsickle
Prediction kinds: proteasome_cleavage
Default scanning lengths: 9
Installation: pip install pepsickle
License: open

NetChop¶

Python classes: NetChop
CLI names: netchop
Prediction kinds: proteasome_cleavage
Default scanning lengths: 9
Installation: your licensed DTU install (NETCHOP_HOME)
License: dtu

NetCleave¶

Python classes: NetCleave, NetCleave_I, NetCleave_II
CLI names: netcleave, netcleave-i, netcleave-ii
Prediction kinds: proteasome_cleavage, endolysosomal_cleavage
Default scanning lengths: 9 (I), 15 (II)
Installation: mhctools fetch netcleave --accept-license
License: unlicensed

DeepTAP¶

Python classes: DeepTAP
CLI names: deeptap
Prediction kinds: tap_transport
Default scanning lengths: (any)
Installation: mhctools fetch deeptap + torch Python
License: open

ERAMER¶

Python classes: ERAMER
CLI names: eramer
Prediction kinds: erap_trimming
Default scanning lengths: (9-16 supported)
Installation: mhctools fetch eramer + openpyxl
License: open

Immunogenicity¶

See the immunogenicity guide for examples and model notes.

Predictor	MHC class	Input
Calis	I	peptides only
PRIME	I	peptides + alleles
DeepImmuno	I	peptides + alleles
TLimmuno2	II	peptides + class II alleles

Calis¶

Python classes: Calis
CLI names: calis
Prediction kinds: immunogenicity
Default scanning lengths: (any)
Installation: built in
License: builtin

PRIME¶

Python classes: PRIME
CLI names: prime
Prediction kinds: immunogenicity
Default scanning lengths: 9
Installation: upstream clone + MixMHCpred 3.0+
License: academic

DeepImmuno¶

Python classes: DeepImmuno
CLI names: deepimmuno
Prediction kinds: immunogenicity
Default scanning lengths: 9-10 only
Installation: mhctools fetch deepimmuno + Keras-2-capable Python
License: open

TLimmuno2¶

Python classes: TLimmuno2
CLI names: tlimmuno2
Prediction kinds: immunogenicity
Default scanning lengths: (set per call)
Installation: mhctools fetch tlimmuno2 --accept-license
License: unlicensed

TCR specificity¶

See the tcr specificity guide for examples and model notes.

Predictor	MHC class	Input
NetTCR	I	(peptide, TCR) pairs
Tulip	I	(peptide, TCR) pairs + `mhc=`
MixTCRpred	I or II (per model)	TCRs against one fixed pMHC target per model

NetTCR¶

Python classes: NetTCR
CLI names: (none)
Prediction kinds: pMHC_TCR_binding
Default scanning lengths: (n/a)
Installation: mhctools fetch nettcr --accept-license + mhctools[nettcr]
License: academic

Tulip¶

Python classes: Tulip
CLI names: (none)
Prediction kinds: pMHC_TCR_binding
Default scanning lengths: (n/a)
Installation: mhctools fetch tulip + isolated Python 3.11
License: open

MixTCRpred¶

Python classes: MixTCRpred
CLI names: (none)
Prediction kinds: pMHC_TCR_binding
Default scanning lengths: (n/a)
Installation: mhctools[mixtcrpred] + mhctools fetch mixtcrpred --accept-license
License: academic

Peptide half-life¶

See the peptide half-life guide for examples and model notes.

Predictor	MHC class	Input
PeptiVerse	none	peptides or `PeptideInput`
PlifePred2	none	peptides only (12-100 residues, natural)

PeptiVerse¶

Python classes: PeptiVerse
CLI names: peptiverse
Prediction kinds: peptide_half_life
Default scanning lengths: (sequence only)
Installation: pinned snapshots (PEPTIVERSE_HOME, PEPTIVERSE_ESM_HOME)
License: open

PlifePred2¶

Python classes: PlifePred2
CLI names: plifepred2
Prediction kinds: peptide_half_life
Default scanning lengths: (12-100)
Installation: plifepred2==1.0 + Pfeature (PLIFEPRED2_HOME, PFEATURE_HOME)
License: open

License tiers¶

Tier	Meaning
open	Open source; code and weights can be fetched or installed freely
academic	Academic / non-commercial; review the upstream license first
dtu	DTU academic license, identity-bound; you request it from DTU
unlicensed	Upstream publishes no license; `--accept-license` records your own authorization
builtin	Built into mhctools; nothing to download

TCR predictors have no mhctools command-line prediction name: their input is a peptide plus a TCR, not an allele. MixTCRpred has its own mhctools mixtcrpred subcommand; see the command line.

Cleavage models (mhctools cleavage) are a separate panel; see cleavage models.