Research blog
Genolux Research Blog
Working notes from the Genolux team: CDR loop prediction methodology, Rosetta energy function behavior, phage display cost breakdowns, AlphaFold2 accuracy limits, and what the developability literature actually says about which physicochemical flags predict failure.
All articles
CDR Design
CDR Optimization and Affinity Maturation: When to Go Computational
Affinity maturation via single-point CDR mutations is one of the most tractable computational problems in antibody engineering — if you know which positions matter.
Affinity Prediction
Predicting Binding Affinity Without an Experimental Structure
Most antibody discovery programs don't have a co-crystal structure at the hit identification stage. We describe how AlphaFold2 Fv predictions substitute for experimental structures.
Developability
Developability Scoring: Which Physicochemical Properties Actually Predict Failure
Dozens of physicochemical properties have been proposed as developability flags. We analyzed 400 therapeutic antibodies to identify which 7 metrics have predictive value at the sequence stage.
Methods Comparison
Phage Display vs. Computational Screening: A Cost-Per-Lead Analysis
A standard phage display campaign screens 109 variants over 6-8 weeks at $600K-$1.2M fully loaded. The math makes the case for computational pre-screening plainly.
Methods
AlphaFold2 for Antibody Design: Real Limitations You Need to Know
AlphaFold2 transformed protein structure prediction — but its CDR loop accuracy, especially for CDR H3 loops longer than 14 residues, is meaningfully worse than for structured domains.
Developability
Manufacturability Assessment in Early Antibody Design: Stop Ignoring Expression Yield
Low-expressing variants routinely make it to cell-line development before the issue surfaces. We describe the sequence features that predict expression failure.
Methods
Rosetta Energy Functions for Antibody-Antigen Docking: REF2015 vs. Talaris2014
Rosetta's antibody-antigen docking performance depends heavily on which energy function you use — REF2015 is not uniformly better than Talaris2014 for all docking scenarios.
CDR Design
Computational Paratope-Epitope Mapping Without Experimental HDX or Cryo-EM
HDX and cryo-EM give definitive epitope maps — at significant cost and time. We describe how contact residue prediction provides actionable paratope-epitope maps for earlier design decisions.
CDR Design
Antibody Humanization: Computational Strategies for Immunogenicity Reduction
Grafting murine CDRs onto human frameworks while maintaining affinity is one of the oldest problems in therapeutic antibody engineering. We review computational scoring approaches.
Affinity Prediction
CDR L3 and H3 Loop Length Diversity in Approved Therapeutic Antibodies
Analysis of 150 FDA-approved therapeutic antibodies from Thera-SAbDab reveals strong biases in CDR L3 (peak at 9 residues) and H3 (distribution 10-16 residues) lengths.
Methods
Benchmarking Antibody Structure Prediction Models on SAbDab: 2025 Update
We benchmarked five antibody-specific structure prediction methods on a 200-structure holdout from SAbDab. Results and evaluation code are released openly.