Help & Documentation

Input formats, result interpretation, and frequently asked questions.

Overview

CPPLocPred is a two-stage machine learning tool for predicting cell-penetrating peptides (CPPs) and their subcellular localization. It accepts peptide or protein sequences in FASTA format and returns:

  • Whether each sequence is a CPP or Non-CPP
  • The predicted intracellular destination for CPPs (Cytoplasm, Nucleus, Mitochondria, Endo/Lysosome, or Others)

A separate Motif Search tool scans sequences for known CPP-associated motifs linked to specific subcellular compartments using the MERCI motif locator.

CPPLocPred uses an ExtraTrees classifier (Stage 1, AAC features) for CPP/Non-CPP classification and CatBoost classifiers (Stage 2, DDR features) for localization prediction.

Input Format

All tools accept sequences in standard FASTA format. Each sequence must have a header line beginning with > followed by the sequence on the next line(s). Multiple sequences can be submitted at once.

>peptide_1 RQIKIWFQNRRMKWKK >peptide_2 GRKKRRQRRRPPQ >peptide_3 NALAALAKKRQIKIW

Accepted Characters

Only the 20 standard amino acid single-letter codes are accepted:

A C D E F G H I K L M N P Q R S T V W Y
Sequences containing non-standard characters (B, J, O, U, X, Z, or symbols) will be flagged as Invalid and excluded from prediction.

Sequence Length

There is no strict length limit. However, CPPLocPred is optimised for short-to-medium peptides (typically 5–50 amino acids). Very long sequences (>500 aa) can be submitted but are not the primary use case.

File Upload

You can either paste sequences directly into the text box or upload a .fasta / .txt file. If both are provided, the uploaded file takes priority and the text box is cleared.

CPP Prediction

The prediction tool runs a two-stage classification pipeline on each submitted sequence.

Stage 1 — CPP vs Non-CPP

  1. Amino acid composition (AAC) features are computed — 20 values representing the percentage of each amino acid in the sequence.
  2. An ExtraTrees classifier predicts the probability of the sequence being a CPP.
  3. If the probability exceeds the selected threshold, the sequence is classified as CPP and passed to Stage 2.

Stage 2 — Subcellular Localization

  1. Distance-based Discrimination Rate (DDR) features are computed — 20 values capturing the distribution pattern of each amino acid along the sequence.
  2. Five independent CatBoost classifiers predict the probability for each location: Cytoplasm, Nucleus, Mitochondria, Endo/Lysosome, Others.
  3. Locations whose probability exceeds their respective optimised threshold are reported. If no location clears its threshold, the highest-scoring location is reported as Uncertain.

Threshold

The CPP threshold (default 0.4) controls sensitivity vs specificity. Lower values increase sensitivity (more CPPs predicted, more false positives). Higher values increase specificity (fewer CPPs predicted, fewer false positives).

Understanding Results

After submission the results page shows a sortable table with one row per sequence. Probability scores are colour-coded for quick scanning:

  • Green (≥ 0.7) — High confidence
  • Amber (0.4 – 0.69) — Moderate confidence
  • Grey (< 0.4) — Low confidence

The Predicted Localization column may contain multiple semicolon-separated locations (e.g. Cytoplasm;Nucleus) if the sequence scores above threshold for more than one compartment.

Results can be downloaded as CSV or TSV using the buttons above the table.

Subcellular Locations

CPPLocPred predicts five destination categories:

LocationDescription
Cytoplasm General cytosolic distribution following membrane translocation. The most common CPP destination.
Nucleus Nuclear-targeted CPPs typically carry NLS-like (nuclear localisation signal) motifs such as polyarginine stretches.
Mitochondria Mitochondria-penetrating peptides often carry MTS (mitochondrial targeting sequence) signals and amphipathic helices.
Endo/Lysosome CPPs that traffic through endosomal or lysosomal compartments after endocytic uptake.
Others Sequences targeting other intracellular compartments (ER, Golgi, plasma membrane, etc.) not covered by the four classes above.

Motif Search

The motif search tool scans your sequences for known CPP-associated motifs using MERCI (Motif EmeRging and with Classes of Importance), a Perl-based motif locator.

How to Use

  1. Paste or upload your sequences in FASTA format.
  2. Select the subcellular location to scan for — this determines which set of motifs is used.
  3. Select a colour class (None, Koolman, Betts-Russell, or RasMol). This selects the motif database variant associated with that amino acid property scheme.
  4. Click Run Motif Search.

Colour Classes

The four colour classes correspond to different motif databases derived from amino acid property groupings:

  • None — Standard motif set, no property-based grouping
  • Koolman — Motifs grouped by Koolman & Röhm biochemical properties
  • Betts-Russell — Motifs grouped by Betts & Russell physicochemical classes
  • RasMol — Motifs grouped by RasMol colour convention (structural/chemical similarity)

Results

The results page shows:

  • A summary DataTable — one row per motif hit across all sequences (searchable and sortable)
  • Per-sequence cards — each sequence displayed with matched residues highlighted in yellow, and a breakdown table of individual hits below
  • Filter tabs to show All / With hits / No hits

Output Columns

Prediction Output (CSV/TSV)

ColumnDescription
IDSequence identifier from the FASTA header
SequenceAmino acid sequence
CPP_ProbabilityStage 1 probability score (0–1). Higher = more likely CPP.
CPP_PredictionCPP, Non-CPP, or Invalid
Cytoplasm_ProbabilityStage 2 localization score for Cytoplasm (only for CPPs)
Nucleus_ProbabilityStage 2 localization score for Nucleus
Mitochondria_ProbabilityStage 2 localization score for Mitochondria
Endo_lysosome_ProbabilityStage 2 localization score for Endo/Lysosome
Others_ProbabilityStage 2 localization score for Others
Final_LocalizationPredicted location(s). NA for Non-CPPs. May be semicolon-separated if multiple locations are predicted.

Motif Search Output (CSV)

ColumnDescription
Sequence_IDSequence identifier
SequenceFull amino acid sequence
Motif_PatternThe MERCI motif pattern that was matched (space-separated residue codes)
StartStart position of the hit (1-based)
EndEnd position of the hit (1-based)
Matched_ResiduesActual amino acid string at the matched position

Standalone Tool

CPPLocPred can be run locally via the command line. Download the standalone package from the Download page.

Requirements

  • Python 3.6 or higher
  • pandas, scikit-learn, catboost
  • Perl (required for motif search only)

Usage — Prediction

python CPPLocPred.py -i input.fasta -o results.csv -j 1 python CPPLocPred.py -i input.fasta -o results.csv -j 1 -t 0.5

Usage — Motif Search

python CPPLocPred.py -i input.fasta -o motifs.csv -j 2 -l Nucleus python CPPLocPred.py -i input.fasta -o motifs.csv -j 2 -l Mitochondria -c Koolman

Arguments

FlagDescriptionDefault
-iInput FASTA filerequired
-oOutput CSV filerequired
-jJob: 1 = Prediction, 2 = Motif Search1
-tCPP probability threshold (Job 1 only)0.44
-mDirectory containing model .pkl files (Job 1 only)./
-lLocation for motif search (Job 2 only)Cytoplasm
-cMotif class: None, Koolman, Betts-Russell, Rasmol (Job 2 only)None
--motif_dirDirectory containing motif files (Job 2 only)./motifs

FAQ

How many sequences can I submit at once?

There is no hard limit, but we recommend submitting up to 500 sequences per job for reasonable response times. For larger datasets, use the standalone tool.

Why does my sequence show as Invalid?

Your sequence contains characters not in the 20 standard amino acid alphabet. Check for spaces, numbers, or non-standard residue codes (B, J, O, U, X, Z). Remove or replace these before resubmitting.

What does "Uncertain" mean in the Final Localization column?

When a CPP is predicted (Stage 1 passes) but none of the five location classifiers (Stage 2) score above their respective thresholds, the highest-scoring location is reported as Uncertain (Best=Location, Prob=X.XXX). This means the sequence is likely a CPP but its localization is not confidently determined.

Can a sequence be predicted to multiple locations?

Yes. If a CPP scores above the threshold for more than one location in Stage 2, all predicted locations are reported, separated by semicolons (e.g. Cytoplasm;Nucleus).

What is the difference between the motif colour classes?

Each colour class (None, Koolman, Betts-Russell, RasMol) uses a different motif database derived from CPP sequences grouped by amino acid property schemes. If you are unsure which to choose, start with None, which uses the general motif set without property-based grouping.

No motifs were found for my sequence — does that mean it is not a CPP?

Not necessarily. The motif search and the prediction tool are independent. A sequence can be predicted as CPP without matching any known motif patterns, and vice versa. We recommend using both tools together for a comprehensive analysis.

How do I cite CPPLocPred?

If you use CPPLocPred in your research, please cite:

Bajiya N., Mehta N.K., Raghava G.P.S. (2025) CPPLocPred: Machine learning-based prediction and subcellular localization of cell-penetrating peptides. IIIT Delhi. Available at: https://webs.iiitd.edu.in/raghava/cpplocpred/