Reference framework
The frozen Core244 reference set contains 31 curated ncMTS/ncMLS-associated references and 213 classical cleavable-presequence references.
HUNTER — Human mitochondrial targeting and N-terminome evidence resource is a computational framework developed by The Goyal Lab to prioritize proteins with sequence features associated with non-classical mitochondrial targeting (ncMTS/ncMLS).
HUNTER is designed as a sequence-architecture prioritization resource. It does not assign mitochondrial localization and does not replace experimental evidence for mitochondrial import or targeting mechanism.
HUNTER evaluates aggregate physicochemical and compositional properties of the protein N terminus using N30, N60 and N100 sequence descriptors. The frozen v1 classifier was developed to distinguish sequence architectures represented by curated non-classical mitochondrial-targeting examples from classical cleavable-presequence architectures.
An above-threshold score therefore identifies a protein for further investigation; it is not a mitochondrial probability.
The frozen Core244 reference set contains 31 curated ncMTS/ncMLS-associated references and 213 classical cleavable-presequence references.
HUNTER v1 uses 42 generic N-terminal sequence descriptors. The stable descriptors include N30 charge proxy per residue, N30 acidic fraction, N60 charge proxy per residue and N30 arginine fraction.
The model and threshold are not recalibrated on the human proteome screen. The frozen decision threshold is 0.542866.
HUNTER Human Proteome Screen v1 (HPS-v1) applies the frozen classifier to the human gene-centric proteome and adds independent interpretation layers without retraining HUNTER. The reference contains 20,484 scoreable protein records corresponding to 20,410 unique normalized amino-acid sequences. The web interface indexes those unique sequences by exact normalized-sequence SHA-256; one sequence may therefore map to more than one HPS-v1 protein record.
Independent context for classical targeting and cleavable targeting sequences.
Sequence-based localization context shown separately from HUNTER.
Residue-wise internal mitochondrial-targeting propensity and peak/window descriptors.
A continuous spatial/localization-context measure reported independently.
HUNTER results are designed for prioritization and biological interpretation, not for a single yes/no mitochondrial-localization decision. The HUNTER score should first be interpreted on its own, and then considered alongside the independent targeting and localization tracks.
The score represents similarity to the sequence-architecture pattern learned by the frozen HUNTER v1 classifier. The frozen decision threshold is 0.542866.
Above threshold: the sequence contains features associated with the ncMTS/ncMLS-like architecture represented by HUNTER v1 and is prioritized for further investigation.
Below threshold: the sequence is not prioritized by the frozen classifier. This does not exclude mitochondrial localization or another targeting mechanism.
The distance from threshold shows how far the HUNTER score lies above or below the frozen cutoff. It is useful for ranking and comparing sequences but should not be interpreted as a calibrated biological probability.
HUNTER reports four stable descriptors to help explain the sequence-level basis of the score:
These descriptors are explanatory model inputs, not independent localization tests.
| Evidence track | What it contributes | How to interpret it with HUNTER |
|---|---|---|
| TargetP 2.0 | Classical N-terminal targeting context, including mTP, signal peptide and noTP probabilities. | A strong mTP or signal-peptide prediction provides an alternative classical explanation for the N terminus and should be considered separately from the HUNTER score. |
| MitoFates | Classical mitochondrial presequence propensity, cleavage-site context, charge and motif information. | Use it to assess whether a protein is better explained by a conventional cleavable presequence rather than a non-classical targeting hypothesis. |
| TPpred3 | Transit-peptide prediction and associated interval/score information. | Use as another independent classical-targeting comparator. It is not merged mathematically with HUNTER. |
| DeepLoc | Sequence-based subcellular-localization context. | Mitochondrial support can strengthen prioritization, but DeepLoc is not experimental evidence and does not establish import mechanism. |
| iMLP | Residue-wise internal mitochondrial-targeting propensity, peak position and best 20-aa window. | Use the profile to identify candidate internal regions for mechanistic follow-up. The iMLP peak is not a HUNTER-defined targeting sequence. |
| PUPS | Continuous spatial/localization-context score. | Use as descriptive context only. PUPS is not converted into a HUNTER score and is not treated as independent experimental evidence. |
This pattern can be useful for prioritizing proteins whose sequence architecture is not readily explained by a conventional cleavable targeting peptide. Additional mitochondrial localization/import evidence remains necessary.
The classical targeting result provides an alternative explanation for the N terminus. The HUNTER result should not be interpreted as evidence for a non-classical mechanism without further support.
Concordant DeepLoc, iMLP or other independent evidence can increase biological interest, but no single combined mitochondrial probability is calculated and concordance is not proof of import.
A below-threshold score means the protein is not prioritized by this frozen sequence-architecture model. It does not rule out mitochondrial localization, classical targeting or mechanisms outside the HUNTER v1 feature space.
These values describe performance on the frozen external sequence-architecture validation panel. They should not be interpreted as accuracy estimates for mitochondrial localization across the human proteome.
The HUNTER v1 model, feature list, scoring engine and threshold are frozen and integrity-checked. Web-interface changes do not alter the scientific model.
Download the frozen H5-I 5+5 example FASTA. The file contains five frozen H5-I non-classical references above threshold and five frozen H5-I classical-presequence references below threshold. It is an illustrative interface example, not an additional validation cohort. Expected scores and calls are provided for verification.
Download the preserved deterministic 10-sequence regression FASTA. This fixture is retained for end-to-end software parity testing and is not a biological validation cohort. Expected scores and calls are provided.
Download HUNTER v1 CLI for Linux x86-64. This is the same frozen executable used by the web server and requires glibc 2.28 or newer. Quick-start instructions · SHA-256 checksum · BSD 3-Clause License.
The web analysis workflow does not persist submitted sequences as part of the HUNTER result system. Temporary batch FASTA files are removed after processing.
Precomputed HPS-v1 interpretation is attached only when the submitted sequence exactly matches a frozen human-proteome sequence by normalized sequence SHA-256.
Software citation: Goyal, P. (2026). HUNTER — Human mitochondrial targeting and N-terminome evidence resource (Version 1.0.1). Zenodo. https://doi.org/10.5281/zenodo.23196383.
Source code and the immutable v1.0.1 release are available at GitHub. The concept DOI for the HUNTER software series is 10.5281/zenodo.23196382. The associated research manuscript is in preparation.
HUNTER v1 software is released under the BSD 3-Clause License. Copyright (c) 2026, Pankaj Goyal.
Third-party predictors, databases and reference resources retain their own licenses, citations and terms of use. Their outputs are presented as independent interpretation tracks and are not incorporated into the frozen HUNTER model.
HUNTER is developed and maintained by The Goyal Lab, A Molecular Medicine Laboratory in the Department of Biotechnology, Central University of Rajasthan.
Visit The Goyal Lab