Score guide

Target and reference sequences

Highlighting follows the recorded targetable intervals, which can describe mutations, junctions or other target regions. It does not establish tumor specificity by itself. A vertical bar marks a zero-length deletion junction. Underlining marks differences only where the recorded comparison supports corresponding positions. Unequal-length pairs are shown without an inferred alignment. The reference label identifies annotation/supplied WT or an exact Isovar patient-background comparison, including its somatic reversion policy. A constructed comparator is not evidence of an observed patient-normal protein. Missing sequences and missing predictions remain distinct.

IC50 (nM): lower indicates stronger predicted binding

Binding affinity from the named predictor. Target and reference values use the same allele, predictor and version. Missing affinity is not zero affinity. Conflicting comparator measurements are labeled rather than reduced to one value.

Vaxrank score

The configured per-allele ranking score. The default transforms predicted binding affinity onto a 0-1 scale, with higher values favored. A custom scoring expression can use other evidence; see the recorded run configuration. This score is shared across predictor rows for the same epitope and allele, not an additional output of each predictor.

Combined score

The configured candidate-window ranking score, combining epitope evidence and, when available and configured, expression evidence. It is distinct from a model's presentation score.

NetMHCpan Score_EL: higher indicates stronger predicted presentation

The eluted-ligand prediction score for this peptide and allele. It is not a measured presentation rate or a probability of vaccine effectiveness. NetMHCpan output definitions.

MHCflurry presentation and processing: higher is stronger

MHCflurry's presentation model combines binding and antigen-processing predictions. Its allele-independent processing score is a separate model output, not a Pepsickle cleavage score. MHCflurry output definitions.

Presentation rank (%): lower is stronger

The predictor's percentile rank relative to its reference distribution. Raw presentation scores from different models are not interchangeable; model and version are retained alongside the values.

Pepsickle cleavage annotations

C-term is the predicted cleavage probability at the epitope's C-terminal bond; max internal is the largest predicted cleavage probability inside it. A source-sequence endpoint has no measured boundary bond and is labeled accordingly. Vaxrank's cleavage composite is sqrt(C-term × (1 - max internal)); it is not a calibrated presentation probability. Pepsickle model description. These annotations do not change ranking. Disabled or missing cleavage evidence means unassessed.