Help
Documentation
What each page and each section of PTMKB shows, how its numbers are obtained and how to read them.
Getting started & search
The search box on the home page finds proteins and modification types. Suggestions appear after two characters; press Enter to open the top match or to see the full list on the Browse page. You can search by:
| UniProt accession | For example P04637. Opens the protein page directly. |
|---|---|
| Isoform | For example O14639-4: accession plus isoform number. |
| UniProt entry name | For example G6PI_HUMAN or VP26A_MOUSE (gene mnemonic, underscore, species code). It is resolved to its accession through UniProt. |
| Gene name | For example TP53. |
| Protein name | Any part of the name, for example tumor suppressor. |
| Modification type | For example Phosphorylation. Opens Browse filtered to proteins carrying that modification. |
If a UniProt entry has been deleted or demerged, the protein page says so and, where UniProt names one, links to the replacing accession. When you are logged in, the proteins you open are recorded in your history.
Browse
Browse lists every protein in the database with its number of modification sites and a count per modification type. The list can be narrowed by:
- Search text: accession, entry name, gene or protein name (at least two characters).
- PTM type: only proteins with at least one site of that modification.
- Curation status: see .
Choose 25, 50 or 100 rows per page and move with the page buttons or the “Go to” box. Filters and the page number are kept in the address, so a filtered view can be bookmarked or shared. Select a row to open the protein page.
Protein page
A protein page is made of the sections below, in this order. The list on the left (a bar of tabs on small screens) jumps to a section and follows your scrolling.
General information
Identity of the protein, mostly from UniProtKB.
| Protein name, gene, organism | As in UniProtKB. |
|---|---|
| UniProt entry name | The mnemonic identifier, for example G6PI_HUMAN; links to UniProt. |
| UniProt accession | The stable identifier used throughout PTMKB; links to UniProt. |
| Sequence length | Number of amino acids of the canonical sequence. |
| Last updated / last sequence update | When the PTMKB record changed, and when UniProt last changed the sequence. |
| Function, subcellular location | Annotation text from UniProtKB. |
| UniProt-annotated modified residues | Modified residues annotated by UniProt, with the enzymes UniProt names for them. |
| Modification sites | Number of site records in PTMKB (all curation statuses), with one chip per modification type. |
| Curation status | Number of sites per status; see the section on curation statuses. |
| Sequence | The full amino-acid sequence. |
Statistics
- Sites: number of modification sites of the protein.
- Unique PTMs: number of different modification types observed on it.
- Unique references: number of distinct literature records (PubMed IDs) supporting the sites.
- PTM × amino-acid table: rows are modification types, columns are residue types, each cell counts the sites. A darker cell means more sites. Click a cell to highlight those residues in the sequence view and in the 3D structure; click it again to clear.
- Bar chart: sites per modification type.
Protein features
Properties computed from the sequence and, where available, from the AlphaFold model.
| Molecular weight, pI, charge at pH 7 | Mass in daltons, isoelectric point and net charge at pH 7 computed from the sequence. |
|---|---|
| Aromaticity | Relative frequency of Phe, Trp and Tyr. |
| GRAVY | Grand average of hydropathy (Kyte–Doolittle). Positive values mean a more hydrophobic protein. |
| Instability index | Guruprasad et al. 1990. Values above 40 predict a protein that is unstable in a test tube. |
| Helix / sheet / turn fraction | Fraction of residues expected in each secondary-structure class, from composition. |
| Mean pLDDT, fraction pLDDT > 70 | AlphaFold's per-residue confidence averaged over the chain, and the share of residues above 70 (generally confident). |
| Total SASA | Solvent-accessible surface area of the model, in Ų. |
| Radius of gyration | Compactness of the model in Å; smaller means more compact. |
| Mean contacts per residue | Average number of neighbouring residues in 3D: a measure of how tightly packed the model is. |
Sequence annotations
A track above the sequence shows where every modification lies. Below it, the sequence view prints the whole protein in blocks of 10 residues with numbering. Modified residues are coloured by modification type; a residue with several modification types shows stacked colours.
- The legend has one toggle button per modification type, all on by default; click a button to hide or show that type. All and None switch every type at once.
- Hover a coloured residue to see its position and modifications; click it to open the site detail panel.
- Download PTM data (JSON) saves every site of the protein.
Site detail panel
| Local window | The 21 residues around the site (10 before, 10 after), coloured by modification type, with the site position. |
|---|---|
| Enzymes | Enzymes recorded for the site, best evidence score first (score in brackets), each linking to UniProt. |
| Upstream enzymes (UniProt) | Enzymes UniProt lists for that modified residue, with links. |
| PTM name and RESID IDs | The modification and its entries in the RESID database (AA numbers). Click an ID to see the entry: image, names, formula, mass, references and a small 3D model. |
| Source | The database or databases the site was taken from (for example dbPTM, UniProt) with a dbPTM link. |
| ECO codes | Evidence and Conclusion Ontology codes giving the type of evidence behind the site (for example ECO:0000269, experimental evidence reported in a publication). |
| PubMed | Evidence identifiers, each linking to the PubMed article. |
| Log Sum / Log-Log Product | The two classic sequence-window scores, explained in . |
| Spatial propensity score | The v2 structure-aware score of this residue for the modification, with its percentile, the three parts of the score and the evidence behind it (see Spatial propensity below). Shown as coming soon when the reference tables of that modification are not built yet. |
JPred Structure
Secondary structure is predicted by JPred (JPred4, Drozdetskiy et al. 2015), a neural-network method working from a multiple sequence alignment. JPred runs as a queued job, so nothing is sent until you press the button. PTMKB then checks every 5 seconds and shows the result when it is ready. A finished result is stored, so the same protein is not submitted twice. JPred accepts sequences up to about 800 residues.
H / E / − | Helix, strand (extended) and coil. |
|---|---|
jnetpred | The final consensus prediction for each residue. |
JNETCONF | Confidence of that prediction, 0 (low) to 9 (high). |
| Burial tracks (JNETSOL…) | Predicted solvent accessibility at several thresholds; B marks a buried residue. |
| Alignment | The sequence alignment JPred used: the query first, then its UniRef90 hits. |
The columns of residues that carry a modification are highlighted, so you can see whether a site lies in a helix, a strand or a loop and how confident the prediction is there.
AF/RCSB Structure
Interactive 3D view (Mol*). The source switch chooses between the AlphaFold model (Jumper et al. 2021) and experimental entries from the RCSB Protein Data Bank. The RCSB list is found from the UniProt accession and shows method, resolution and chains, best resolution first.
- Residue mapping. PDB files number residues in their own way, so UniProt positions are mapped to PDB residues with PDBe SIFTS (Velankar et al. 2013). Sites then land on the correct residue and chain.
- Colour modes. pLDDT, secondary structure, chain, or PTM only.
- Representation. Cartoon, ball-and-stick or surface.
- PTM filter and labels. Show one modification type or all, switch labels on or off, and centre the view on a site. Clicking a site in the sequence view or table also selects it here.
- Download DSSP/SASA (JSON). Per-residue simplified and detailed DSSP secondary structure (Kabsch & Sander 1983) and solvent-accessible surface area (Shrake–Rupley, in nm²) for the displayed structure.
Spatial propensity
Spatial propensity is the new v2 scoring mechanism. It uses sequence and 3D structure to score how well each eligible residue of the protein fits the profile of known sites of one modification, so unannotated residues can be flagged as having the potential to be modified. Pick the modification above the plot; known sites are shown in orange. Switch between percentile and score (nats). The score is the sum of three parts:
| Sequence | Neighbouring residues (−10…+10) under this PTM's positional frequency matrix, compared with unannotated residues of the same type. |
|---|---|
| Structural state | Disorder, burial and secondary structure of the site in the AlphaFold model, weighted by pLDDT. |
| 3D neighbourhood | Residues within 12 Å in space but far apart in sequence, weighted by AlphaFold's predicted aligned error and by how confidently the site is ordered. |
- Score. Each part is a log-likelihood ratio (modified versus unmodified residues of the same type), in natural-log units called nats; the total is their exact sum, with no fitted weights. Positive favours the modification, negative argues against it.
- Percentile. Share of unmodified residues of the same type that score strictly lower. A residue at the 95th percentile scores higher than 95% of them. It is calculated from the full score when a structure is available, otherwise from the sequence part only (shown as the percentile basis).
- Evidence panel. For a residue: pLDDT, the probability the site is ordered, relative solvent accessibility (RSA, exposed from 25%), the secondary structure, and the neighbouring residues that contribute most to the 3D part.
no_structure | No AlphaFold model was found for this protein: only the sequence part is scored. The structure-based part will be added soon. |
|---|---|
no_pae_hard_shells | No PAE file: 3D shells use hard distance cut-offs. |
structure_sequence_mismatch | AlphaFold sequence differs from the UniProt sequence at some positions. |
low_confidence_site | Low AlphaFold confidence at this site (pLDDT < 50): 3D evidence is down-weighted. |
window_truncated | Window runs past the protein end or over non-standard residues. |
structure_length_mismatch | The AlphaFold model does not line up with this sequence: only the sequence part is scored. The structure-based part will be added soon. |
site_not_in_model | This residue is outside the AlphaFold model (partial or isoform entry): only the sequence part is scored. |
structure_tables_underpowered | Too few known sites with a structure for this PTM and residue: structural parts are set to 0. They will be added soon, once more data is available. |
If the reference tables of a modification are not built yet, the section says “coming soon” instead of a score. The scores can be downloaded as CSV.
Curation statuses
Every site carries a status describing how it was established.
Experimentally-Verified | The site is backed by experimental evidence on record. |
|---|---|
Consensus-Predictedshown as Consensus-Validated in some filters | The site rests on a consensus across sources or predictors rather than on direct experimental evidence for it. |
UniProt-Predictedshown as UniProt-Validated in some filters | The site comes from a UniProt annotation without direct experimental evidence in PTMKB. |
Validation-Pending | Not yet checked. Hidden by default on the protein page; tick it in the status filter to include these sites. |
Note on naming: the database stores the second and third status as Consensus-Predicted and UniProt-Predicted, while the website's filter code currently uses Consensus-Validated and UniProt-Validated. They denote the same two classes; until the two are unified, one of the spellings may not match in a filter.
Propensity calculator
The Propensity page scores a short sequence window from the original PTMKB positional matrices: how often does each amino acid occur at each position from −10 to +10 around known sites of a modification?
- Subsequence. 13 to 21 residues, odd length, one-letter codes of the 20 standard amino acids. The middle residue is the one being scored (shown in red). The counter shows the length.
- Modification. Only modifications that occur on the middle residue are selectable; the rest are greyed out. Examples fill the form in one click.
- Vector. For each position, the log-e value of the residue found there. −∞ means no known site has that residue at that position.
- Log Sum. The sum of all values of the vector. It is −∞ as soon as one value is −∞.
- Log-Log Product. Take the longest stretch centred on the middle residue that has no −∞ (at least 13 residues; otherwise the result is NIL). Multiply its values except the middle one, and return −ln of the absolute product. NIL is also returned if a value in that stretch is 0.
- Table. The matrix the vector came from; the cells matching your sequence are outlined. Blue cells are more frequent than typical, red less.
The positional matrix viewer at the bottom of the page shows any modification/residue matrix as a log-e or freq table. These two scores use the window only; for scores that also use 3D structure, see Spatial propensity on a protein page.
Integration, API & account
Browsing never needs an account. An account is needed for the REST API.
- Sign up / log in on the Login page. A username has 8–32 letters, digits or underscores; a password at least 8 characters. No email address is stored. You get a personal API token at once.
- User menu. After login, your name opens a menu with Profile, Copy API token and Logout.
- Token. Valid for 30 days. Send it as
Authorization: Bearer <token>. Your profile shows it masked, with Copy, Reset (the old token stops working at once) and the time left. An expired token has to be reset. - Daily allowance. Each user has a number of API calls per UTC day (20,000 by default). Today's use and the last 30 days are on the profile page. Calls above the allowance get HTTP 429 until 00:00 UTC.
- History. The profile lists your website searches and API calls, filterable by Web or API, with links to the proteins.
- Integration page. Integration has the token, the starter script
ptmkb_get_started.py, curl and Python examples, and interactive API documentation (Swagger UI) with your token already entered. - Old scripts. The PTMKB v1 endpoints under
/ptmkb/api/keep working with their old response format; they need a v2 token and use the same allowance.
Downloads
The Download page offers snapshots of the database and positional matrices for local analysis. To look at a single matrix without downloading it, use the matrix viewer on the Propensity page.
References
- Jumper J, et al. Highly accurate protein structure prediction with AlphaFold. Nature 596:583–589 (2021).
- Drozdetskiy A, Cole C, Procter J, Barton GJ. JPred4: a protein secondary structure prediction server. Nucleic Acids Res 43:W389–W394 (2015).
- Velankar S, et al. SIFTS: Structure Integration with Function, Taxonomy and Sequences resource. Nucleic Acids Res 41:D483–D489 (2013).
- Kabsch W, Sander C. Dictionary of protein secondary structure: pattern recognition of hydrogen-bonded and geometrical features. Biopolymers 22:2577–2637 (1983).
- Shrake A, Rupley JA. Environment and exposure to solvent of protein atoms. Lysozyme and insulin. J Mol Biol 79:351–371 (1973).
- Garavelli JS. The RESID Database of Protein Modifications as a resource and annotation tool. Proteomics 4:1527–1533 (2004).
- Guruprasad K, Reddy BVB, Pandit MW. Correlation between stability of a protein and its dipeptide composition. Protein Eng 4:155–161 (1990).
- Good IJ. Probability and the Weighing of Evidence. Griffin, London (1950). The basis for adding log-likelihood ratios as in the v2 propensity.