Using B cell receptor lineage structures to predict affinity
Fig 2
Simulation performance for Δ-affinity prediction vs. observation time for the metrics in Table 1.
The y-axis describes how close each metric comes to achieving the best possible performance. Here we are predicting the location of affinity-increasing mutations in a lineage of inferred ancestral sequences, and for example an “accuracy gap” value of 0.5 for nuc-lbr would mean that if we choose the best inferred ancestral sequence using nuc-lbr that we would, on average, be 0.5 branches away from a branch containing an actual affinity-increasing mutation. See Fig 1 caption for details. Similar plots across ranges of other simulation parameters can be found at https://zenodo.org/record/3929565.