cdskit localize targeting5 v1 - kfuku52/cdskit GitHub Wiki
cdskit localize targeting5-v1
The CDSKIT-trained weights are MIT-licensed; the model release includes LICENSE and SHA256SUMS. The license update leaves the checkpoint bytes and legacy runtime requirements unchanged. Source datasets retain their original terms.
targeting5-v1 is the pretrained model used by:
cdskit localize --seq_file proteins.faa --seq_type protein --model targeting5 --report localize.tsv
It predicts the TargetP-compatible five-class targeting-peptide labels:
| TargetP label | cdskit label |
|---|---|
Other |
noTP |
SP |
SP |
MT |
mTP |
CH |
cTP |
TH |
lTP |
The model file is cdskit-localize-targeting5-v1.pt. When it is not already
cached, cdskit localize --model targeting5 downloads it from the cdskit
GitHub Release and verifies its SHA-256 checksum before loading it.
Dependencies
The pretrained model runs on CPU, but it uses optional ML dependencies at runtime:
torchscikit-learn
The embedded sklearn estimators were serialized with scikit-learn 1.5.2. Use the dedicated runtime setup; current CDSKIT supports their known legacy loss layout with scikit-learn 1.9.0 and has been compared with the 1.5.2 reference. The alias verifies the release checksum before enabling legacy pickle loading.
Model cache
The cache root is $CDSKIT_MODEL_DIR, else $XDG_CACHE_HOME/cdskit/models,
else ~/.cache/cdskit/models. This artifact is stored under
localize/targeting5-v1/v1/cdskit-localize-targeting5-v1.pt within that root.
Its SHA-256 is
ddaeab7093533a213ee58117b70ad0f45b0c126cf82c77df32e369eaff2beeb2.
See offline and safety notes.
Training data
TargetP-2.0 and the cdskit targeting5-v1 model use the same five target
labels, but they do not use identical training recipes. cdskit does not use
TargetP executable code, trained weights, or fitted parameters.
| Data source | Rows | Class counts | Role |
|---|---|---|---|
TargetP-2.0 public table, targetp2_benchmark.tsv |
13,005 | noTP 9,537; SP 2,697; mTP 499; cTP 227; lTP 45 | canonical TargetP-compatible supervised labels for base training |
Strict external table, targetp2_external_torch_strict_cleanholdout_seed1seed2_plus_thylum_h128_e6_external.tsv |
13,983 | noTP 4,684; SP 3,986; mTP 2,914; cTP 1,872; lTP 527 | external augmentation for base PyTorch models |
| Reranker development holdout | 20,000 | 4,000 per class | trains the multiclass HGB reranker and class thresholds |
| mTP/noTP specialist train | 16,000 | mTP 8,000; noTP 8,000 | trains the binary mTP/noTP specialist |
| mTP/noTP specialist validation | 2,000 | mTP 1,000; noTP 1,000 | selects the specialist threshold, currently 0.41 |
Model structure
| Component | cdskit targeting5-v1 |
|---|---|
| Base predictors | Two cdskit-trained TargetP2-style PyTorch models (sqrt and logcw) |
| Blend | Classwise alpha 0.4 |
| Thresholds | noTP 1.0, SP 0.65, mTP 1.0, cTP 0.8, lTP 1.0 |
| Reranker | scikit-learn HGB multiclass reranker |
| Specialist | scikit-learn HGB mTP/noTP specialist |
| Inference device | CPU |
Performance snapshot
These are historical development results associated with the published model, not an official TargetP-2.0 server rerun or a new evaluation of the current CDSKIT version. The TargetP paper row and the CDSKIT holdouts use different datasets, class distributions, and evaluation protocols.
| Model / evaluation | Rows | Macro F1 | Accuracy | noTP F1 | SP F1 | mTP F1 | cTP F1 | lTP F1 |
|---|---|---|---|---|---|---|---|---|
| TargetP-2.0 paper Table 1 | - | 0.890 | - | 0.980 | 0.980 | 0.860 | 0.880 | 0.750 |
| cdskit targeting5-v1, seed2 strict external holdout | 1,801 | 0.891 | 0.895 | 0.773 | 0.921 | 0.851 | 0.950 | 0.957 |
| cdskit targeting5-v1, fresh unused balanced holdout | 1,000 | 0.907 | 0.909 | 0.796 | 0.915 | 0.861 | 0.976 | 0.990 |
These numbers do not establish equivalent accuracy to TargetP-2.0. That claim
would require both predictors to be evaluated on the same independently held-out
data with a compatible protocol. This model predicts targeting-peptide classes,
not general DeepLoc localization labels; its original p_peroxisome head is a
constant zero, not a trained peroxisome detector.
Sources
Feature and decision versions
See the localization feature and decision contract
for the corrected nine-residue PTS2 definition, legacy model compatibility,
--decision_policy safe-v1, v2 output and explicit taxonomy constraints.
Published artifacts retain their historical defaults; new staged pipeline models
default to safe inference with ensure_one_label=False. Scores remain unverified
as calibrated biological probabilities.
References
Cite the method or resource actually used, in addition to the CDSKIT version and command. Papers below do not validate newly trained CDSKIT models or new options.
- Almagro Armenteros JJ et al. (2019). Detecting sequence signals in targeting peptides using deep learning. Life Science Alliance 2:e201900429. Paper — TargetP 2.0 methodology and targeting-peptide dataset; CDSKIT targeting5 weights are separately trained, not the official TargetP predictor.
- The UniProt Consortium (2025). UniProt: the Universal Protein Knowledgebase in 2025. Nucleic Acids Research 53:D609–D617. Paper — Protein sequences and annotations; also report the downloaded snapshot/query and evidence filters.