Reproduce “Preference-based Antibody Expression Ranking: Scaling with Large-scale Weak Supervision” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Josh Qixuan Sun, Morteza Babaie, Wenyang Hou, Mark Crowley, David Young
Researchers from Kisoji Biotechnology Inc. and the University of Waterloo developed a preference-based framework that integrates scarce quantitative expression data with millions of weakly supervised antibody sequences to accurately rank antibody expression. This two-stage approach, combining continual pretraining and preference fine-tuning, consistently outperformed supervised regression baselines in both ranking and binary classification, demonstrating scalability with weak supervision.
Key insights
- Large-scale weak positive supervision, derived from empirically expressible sequences, can serve as a robust biophysical regularizer when combined with strong supervision, improving generalization and stability.
- Continual pretraining of pLMs on domain-specific weak data creates a foundational representation space that synergistically boosts the performance of subsequent preference optimization.
- Direct Preference Optimization (DPO) can be efficiently adapted for pLMs, even without tractable autoregressive likelihoods, by using pseudo-log-likelihood approximations and union-masked computations for preference pair differences.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.