easy PESE · PROTEIN EMBEDDING SPACE EXPLORATION made simple
Explore Embedding Space
Discover residue patterns within protein families to uncover sequence diversity
Discover the protein sequence landscape
Explore protein diversity from sequence to structure. PESE combines ESM-2 protein language model embeddings, interactive UMAP projections and cosine-distance trees to reveal relationships within protein families. Start with Pfam, InterPro, a protein sequence or accession, or your own FASTA, PDB or mmCIF data. Discover homologs by BLAST and focus datasets using domain architecture selection.
Compare groups down to individual residues. Explore globally mapped FAMSA2 alignments, group consensus sequences and conservation. Inspect physicochemical properties and ESM-2 negative log-likelihood (NLL) patterns, highlight individual proteins, and map selected positions onto experimental or predicted structures in synchronized 3D viewers.
From exploration to testable hypotheses. Move between family-wide patterns, embedding groups and individual sequences. Export alignments, embedding trees, figures, source annotations and restorable sessions to support comparative analysis and enzyme engineering. Embedding similarity complements conventional sequence analysis but does not, by itself, establish evolutionary relationships or biochemical function.
Resources
PESE is based on amino acid cluster analysis (A2CA) and PLMclust:
- A2CA: D. Eggerichs and D. Tischler (2023), Science Data Bank. DOI: 10.57760/sciencedb.09549. Explore the original amino acid cluster analysis platform at a2ca.app.
- PLMclust: M. Penner et al. (2026), Unsupervised protein language models learn patterns of enzyme function, bioRxiv. DOI: 10.64898/2026.04.23.720319. See plmclust.co.uk.
PESE source code and version history are available in the GitHub repository.
Software and scientific resources
- ESM-2 650M protein language model for sequence embeddings, with GPU inference on Modal and job orchestration on Railway.
- UMAP for interactive visualization of protein embedding space; iterative embedding-based clustering and a hierarchical embedding-distance tree for exploring protein groups.
- FAMSA2 for multiple-sequence alignment and comparisons of residue conservation and group consensus sequences.
- InterPro / Pfam for family annotations and protein datasets; EMBL-EBI HMMER for domain scans against Pfam profiles.
- NCBI BLASTP and NCBI Protein EFetch for optional homolog searches and retrieval of complete protein sequences.
- RCSB Protein Data Bank for protein structures and 3Dmol.js for interactive structure visualization and residue mapping.