Figure 1.

Overview of computational counterselection strategy and experimental validation

(A) Using enrichment over rounds from single-target phage panning as a regression label, we train multi-task ensemble models that jointly predict affinity to on and off targets. We can then use this affinity prediction to identify sequences that bind to the off-target molecule and remove these sequences.

(B) Comparison with molecular counterselection for validation. To compare to molecular counterselection, cross-panning experiments of the on and off target and individual binding assays of 48 selected sequences were done.