AbPACER: parent-aware, affinity-label-blind prioritization of affinity-matured scFv clones from phage-display NGS
Background: Affinity-maturation phage-display next-generation sequencing (NGS) yields more paired single-chain variable fragment clones than can be characterized experimentally, creating a fixed-budget prioritization problem. Read counts provide empirical support rather than direct affinity labels. We developed AbPACER (Antibody Parent-Aware Contextual Evidence Ranker), an affinity-label-blind…
AbPACER is a novel neural ranker designed to prioritize affinity-matured scFv clones from phage-display NGS. This ranker addresses the challenge of analyzing a vast number of candidate clones that cannot all be characterized experimentally due to fixed-budget limitations. AbPACER combines parent-relative mutation descriptors, frozen antibody-language-model context, and read counts from related clones to create a more informed ranking system.
The developers of AbPACER, named Antisera, conducted retrospective evaluations using two phage-display campaigns, targeting two distinct protein families: Fas-associated factor 1 (FAF1) and vascular endothelial growth factor receptor (VEGFR). Each campaign had its own set of frozen top-5% candidate sets, containing 16,323 FAF1 and 7,487 VEGFR clones. AbPACER was set to select 384 candidates for assay lists.
When comparing AbPACER's performance to other learned methods, it demonstrated superior results. In the FAF1 campaign, AbPACER recovered 2.00 +/- 0.00 of the seven retrospective panel clones, achieving the highest recovery rate among the methods tested. This included recovering two out of every five seeds. In contrast, total count recovery was only 1/7, Ens-Grad CNN recovered 1.00 +/- 0.00, A2Binder-HL recovered 1.67 +/- 1.15, and AbAffinity recovered 1.33 +/- 0.58.
Similarly, in the VEGFR campaign, AbPACER-MSE (a variant of AbPACER) recovered 2.33 +/- 0.58 of the three panel clones, marking the highest mean recovery among learned methods. Total count recovery still outperformed AbPACER-MSE at 3/3, but AbPACER-MSE still outperformed other learned methods such as A2Binder (175.7 +/- 6.4) and Ens-Grad CNN (154.0 +/- 6.1) in the public AlphaSeq common split of 11,670 fixed-test variants.
The AbPACER-MSE model was able to recover 187.0 +/- 2.6 of the true top-384 targets, closely matching the performance of AbAffinity (188.0 +/- 2.6). Importantly, AbPACER-MSE utilized a significantly smaller number of task-specific parameters (1.378 million) compared to AbAffinity (651.04 million), making it a more efficient solution for fixed-budget prioritization in phage-display NGS data.
In conclusion, AbPACER has emerged as a promising framework for fixed-budget prioritization in affinity-label-blind phage-display NGS data. By incorporating parent-relative mutation descriptors, language-model context, and NGS evidence, AbPACER provides a campaign-adaptive approach that outperforms other learned methods in terms of recovery rates and parameter efficiency.
Further prospective evaluations of sequence-conditioned reranking could complement count-based prioritization strategies, offering a more robust solution for prioritizing affinity-matured scFv clones.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
