Harmony 2.0 - Accelerating structure-based epitope mapping via a hybrid AI model for de novo protein structure prediction integrating cryoEM data

Short description of the innovation study

Challenge: Antibody–antigen structures accelerate the development of novel, better therapeutics. Current AI prediction models for such complexes remain inaccurate (max ~40% success rate), so experimental structure determination is still the gold standard — yet too costly for broad adoption. Solution: We will enrich existing antigen–antibody complex prediction models with additional inputs, including interaction data and/or low-resolution structural information. These data act as constraints guiding model building, fine-tuning prediction accuracy. We aim to boost predictive strength by at least 40% (>80% success rate), approaching experimental confidence while cutting time and cost per structure ~10-fold. This will make structure-guided antibody development and engineering accessible to all biotech and pharma companies, enabling early selection of the best antibodies from mAb panels.

Organisations involved: 

Main SME: PUXANO