Virtual Protein Universe
Explore the theoretical space of protein sequences. Separate what can be calculated, what databases report, and what still requires experiments.
About this project
Description. Virtual Protein Universe is an interactive, browser-based workbench for exploring protein sequences. Inspired by the Library of Babel, it starts from the idea that the number of possible amino-acid sequences is astronomically larger than anything nature has produced, and it lets you measure that space, analyse individual sequences, look up known proteins, compare variants and view predicted structures.
Aim.
- Make the scale of protein sequence space tangible and teach basic sequence properties (molecular weight, pI, hydropathy).
- Keep four kinds of information visibly separate: what is calculated here, what a database reports, what a model predicts, and what has been shown by experiment.
- Provide a credible foundation for responsible protein-exploration tools, where every feature states what it can and cannot show.
What this is not: it is not a discovery engine, not a diagnostic tool, and nothing here is experimental evidence. A valid sequence is not proof that a protein folds, works or is safe. Educational and exploratory use only; not for diagnosis, treatment or clinical decisions.
What you can do here
- Sequence Analyzer: calculate sequence-space size, then molecular weight, pI and hydropathy of a sequence, see the evidence levels, and show a structure (AlphaFold DB by accession, or your own prediction server).
- Protein Universe: open hypothetical sequences by address (every sequence of a given length has one), draw sequences with chosen composition, browse real UniProt records, save collections and export FASTA or CSV.
- Mutation Lab: compare two sequences and align them to find substitutions, insertions and deletions.
- Database Search: search UniProt records by name, gene, organism or function.
Roadmap (planned, not yet available)
None of these exist in this page yet, and nothing here simulates them. They are listed so no mock-up is mistaken for a working feature. Several need external services or a server.
- Sequence-similarity search (BLAST) against real databases, with identity, coverage and E-value.
- Mutation laboratory with insertions, deletions, property changes and protein language-model scoring (for example ESM-2).
- Structure explorer: experimental PDB structures, structure comparison, secondary structure, PAE and ligands.
- Conservation analysis: multiple sequence alignment, conservation scores, sequence logos and trees from established tools.
- Evidence integration with NCBI, PDB, ClinVar, InterPro, Pfam and Europe PMC, each result with source, record identifier and retrieval date.
- Knowledge Archive: literature search (PubMed and Europe PMC) linked to proteins and mutations.
- Research workspace with notes and PDF reports, and an educational laboratory.
- Browsing the Protein Universe by family, cellular location, pathway and disease, and similarity networks built from real alignments.
- Recording of real laboratory results, which is what Level 3 of the evidence panel would require.
Sequence space calculator
Possible sequences of exactly N residues using the 20 standard amino acids (20N)
1.27 × 10130
This is the theoretical sequence count, not the number of proteins known to exist or fold. It counts only sequences of length exactly N, and it counts every distinct string, including the overwhelming majority that would never fold or function.
Sequence analyzer
Calculated properties
Length (residues)
Molecular weight (average)
Theoretical pI
Average hydropathy (GRAVY)
Amino-acid composition
- Molecular weight: sum of average residue masses plus one water (18.015 Da), i.e. the unmodified, neutral polypeptide. Not a measured mass; excludes the initiator-Met removal, signal peptides, disulfides, glycans and other modifications.
- Theoretical pI: pH at which the net charge is zero, using the EMBOSS pK set, free N- and C-termini and no local-environment effects. Other published pK sets shift the result by several tenths of a pH unit, so treat it as an estimate, not a measurement.
- GRAVY: arithmetic mean of Kyte-Doolittle hydropathy values over all residues (positive = more hydrophobic overall).
Predict a 3D structure for this sequence (optional, needs a prediction server)
Runs ESMFold on a GPU server that you operate (see the GPU backend zip). Paste the server's web address below. The analyzed sequence is sent to that server, so only use a server you trust. Works for any sequence of 16 residues or more, up to the server's limit. The result is a computational prediction, not experimental evidence, and random or designed sequences often give low-confidence, unstructured models.
3D structure from AlphaFold DB (optional)
The AlphaFold Protein Structure Database holds precomputed predicted models for proteins in UniProt. It cannot predict a structure for a new or designed sequence. Enter a UniProt accession (for example P68871; use the search below to find one). If the database sequence is identical to the sequence you analyzed, the Level 2 card is updated.
Clicking this sends the accession (not your sequence) to alphafold.ebi.ac.uk.
Evidence and validation
The cards below report evidence actually available in this prototype. A valid amino-acid string is not proof that a protein folds, works, or is medically useful.
Level 1: Sequence syntax
No sequence analyzed.
Level 2: Structural plausibility
No structure shown yet. Analyze a sequence, then predict one with your prediction server, or enter its UniProt accession. A model is computational, not experimental evidence.
Level 3: Experimental function
No laboratory assay data are collected by this tool. Function is not experimentally verified.
Level 4: Medical utility
No clinical safety, efficacy, pharmacokinetic, or regulatory assessment is performed.
THE PROTEIN UNIVERSE
An exploration of known biology, hypothetical proteins, and the infinite possibilities of molecular sequences.
Two kinds of sequence live here, and every card says which one you are looking at.
- Hypothetical sequences are computed on demand from an address. Nothing is stored. Every possible sequence of a given length has exactly one address, but almost none of them occur in nature, and none is known to fold or function.
- Known sequences are real records fetched from UniProt, with a link back to the record.
Choose a length. Every sequence of that length can be reached, and any sequence you type has an address. Neighbouring addresses give unrelated sequences, by design, like the Library of Babel.
Collections and property map
Property map
Each dot is a sequence you opened this session or saved in this collection. Position is calculated from the sequence itself (horizontal: GRAVY hydropathy; vertical: theoretical pI); dot size grows with length. It shows properties only. It does not show evolutionary relationship or similarity. Click a dot to open its card.
Mutation comparison
For equal-length sequences, compare residues position by position. "Compare" reports substitutions only and needs equal lengths. "Align" runs a global pairwise alignment (Needleman-Wunsch, simple scoring: match +2, mismatch -1, gap -2, up to 2,000 residues each) and reports substitutions, insertions and deletions. Positions are counted from the first residue you enter (so an initiator Met counts as position 1).
UniProt database search
Search known protein records by protein name, gene, organism, or function keywords. This is a text/database search, not a sequence-similarity search.