Protein Evolutionary Pathway Analyser

Works for any protein — SARS-CoV-2 spike, Influenza HA, HIV gp120, Dengue E, or your own.

Two filters combined:

ESM-2 attention asymmetry ΔAsym — which positions gain epistatic pressure after a mutation
Codon transition score C — which amino acid changes are reachable by a single nucleotide substitution

Joint score = ΔAsym × C — positions satisfying both conditions.

Select protein / preset sequence

Choose a built-in example or 'Custom' to paste your own

Mutations to analyse

Comma-separated. The app auto-corrects if the reference AA doesn't match.

Mutation spectrum (codon transition weights)

Select the pathogen's RNA polymerase bias. Use Influenza for H1N1/H3N2/H5N1. Use SARS-CoV-2 for coronaviruses.

Target position sets for comparison (optional)

One set per line: 'Name: pos1, pos2, pos3'
Gained positions will be tested for enrichment in each set (hypergeometric p-value). Leave blank to skip comparison.

Status

Download all data (ZIP)

Left: ΔAsym (attention only). Right: Joint score (attention × codon). Purple = position in a target set. Gold line = mutation site.

Plot

How target sets work: Define any named groups of positions (e.g. known variant mutations, functional sites, antibody epitopes). The app counts how many pathway-opened positions fall in each set and computes a hypergeometric p-value. No external files needed — type the positions directly.

Codon score: Normalised row probability in the weighted amino acid transition matrix, derived from the RNA codon table with SARS-CoV-2 mutation spectrum weights (C→U dominant). Applies to any RNA virus protein. For DNA viruses or proteins under different mutational pressures, the weights can be adjusted in the source code.

Window size: Up to 1022 residues per forward pass (ESM-2 architectural limit). For proteins shorter than 1022 residues, the entire sequence is processed in one pass. For longer proteins, the window is centred on the mutation site.