Enter the HLA class I epitope your TCR-T targets and every position of all 20,000 human proteins is searched for similar peptides the same TCR might cross-recognise, with the expression of each hit’s gene across 54 normal tissues. This is the standard safety step before a TCR goes into in vivo work. Everything is computed locally in your browser; no sequence is uploaded.
1 · Epitope and search criteria
Enter the epitope, then choose what counts as “similar”. The default, “≤2 mismatches”, is for a quick first look; a formal safety assessment should define the motif from alanine-scanning data.
Example:
Per-position criteria
2 · Results
Similar peptide
Mismatches
Gene
Position
Anchor
BLOSUM62
Peak normal-tissue expression
No hits under the current filters.
Method, evidence and limits
Why this step matters
MAGE-A3 / HLA-A*01:01 (2013): an affinity-enhanced TCR cross-recognised a similar peptide in titin, and two patients died of cardiogenic shock. That peptide differs from the target epitope at several positions — only a motif that fixes the critical positions and opens the rest, scanned against the whole proteome, would have found it.
MAGE-A3 / HLA-A*02:01 (2013): the TCR cross-recognised a homologous MAGE-A12 peptide expressed in brain, and two patients suffered fatal neurotoxicity. The off-target protein belongs to the same gene family as the target and is highly similar in sequence.
What both cases share: they had been screened against tumour cell lines in vitro and had partial tissue cross-reactivity data, and still missed it. A whole-proteome scan at the sequence level is the cheapest filter, and the earliest one you can apply.
Which of the three criteria to use
Mismatch count: any N positions may differ. Good for a first pass and for comparing candidates; the larger N, the more hits. Alanine-scan motif (X-scan): substitute each position with alanine and measure function. Positions where function drops sharply are the ones the TCR reads and must be held fixed; the rest are opened to any residue, and that motif is scanned against the whole proteome. This is the standard established after the MAGE-A3 deaths, and the criterion this tool was built around — set the critical positions from your own data to “must match” in the grid above. Conservative substitution: each position allows any substitution scoring ≥ 1 in BLOSUM62. It sits between the other two, giving something closer to the underlying biochemistry than a plain mismatch count without needing experimental data.
Limits you need to know
What this tool does is enumerate candidate peptides at the sequence level. It is not a binding prediction and it is not a safety conclusion. It cannot replace pHLA binding validation, cell-level killing and activation assays, or normal-tissue cross-reactivity testing. Specifically: ① it covers only the 20,000 curated canonical sequences of UniProtKB/Swiss-Prot — no splice isoforms, non-canonical ORFs, post-translationally modified products or individual polymorphism; ② whether a hit is a real epitope also depends on proteasomal processing, TAP transport and HLA binding; this tool only reports agreement at the main anchor positions of the selected HLA and makes no affinity prediction; ③ GTEx is median mRNA expression in normal tissue, which is not the same thing as protein level, cell-subset composition or HLA expression — a low TPM does not mean safe; ④ a tissue reading zero can still be upregulated by inflammation or stress.
Sources
Proteome: UniProtKB/Swiss-Protcurated Homo sapiens entries, used under CC BY 4.0. Tissue expression: GTEx Portal v8 median gene TPM per tissue (RNA-seq, summary level only). Anchor preferences: compiled from the main anchor residues reported for each allele by SYFPEITHI and the MHC Motif Atlas; for guidance only, not a binding prediction. Scoring matrix: BLOSUM62 (Henikoff & Henikoff, 1992).
TCRshows
Cell Therapy knowledge platform