Research & Analyses
Comprehensive analyses of de novo antibody design platforms and developability prediction.
Dogfooding rfab-harness on 10 Challenging Cancer Targets
4,085 De Novo Antibody Designs Scored Across Pediatric & Adult Cancers
First systematic stress-test of rfab-harness across 10 cancer driver targets (B7-H3, GD2, EGFRvIII, HER2, CEACAM5, mesothelin, EphA2, GFAP, CD276, PDGFRA). Full RFdiffusion → ProteinMPNN → RF2 pipeline on Modal A100 GPUs. 135 of 4,085 designs passed stringent RF2 quality filters (pAE < 10, CDR RMSD < 2Å). Pass rates ranged 60-fold across targets—from 0.3% (EGFRvIII) to 19.8% (CEACAM5)—revealing strong target-dependent variation in designability.
rfab-harness
Campaign Orchestration for RFAntibody
One YAML config, one CLI command, full antibody design campaign. Wraps the 3-stage pipeline (RFdiffusion, ProteinMPNN, RF2) with target prep, multi-GPU parallelization, filtering, ranking, and 21 pre-built configs for cancer and rare disease targets.
Does Developability Come For Free?
FLAb Dataset Analysis
Testing whether de novo antibody design models learn therapeutic-like developability without explicit training. Analysis of 160 FLAb datasets reveals training data bias as the explanation for "emergent" developability.
DADB-v1.0: A Therapeutic Decathlon
De Novo Antibody Design Benchmark
The first comprehensive benchmark measuring what actually matters for therapeutic antibodies: binding, structure, developability, and immunogenicity. A composite scoring system for the field.
Antibodies from Thin Air
Five AI Platforms Rewriting Cancer Drug Discovery
A comprehensive comparison of JAM-2, Chai-2, Origin-1, RFAntibody, and Latent-X2. Covers hit rates, binding affinities, oncology targets, and the first immunogenicity data for AI-designed antibodies. Includes a primer on antibody biology for newcomers.