BioPhi

BioPhi automates humanization and humanness evaluation of mouse-derived antibody sequences to support therapeutic antibody development.

Key Features:

• Sapiens: A deep learning language-model-based humanization method trained on the Observed Antibody Space (OAS) to generate human-like antibody sequences at scale and benchmarked on 177 antibodies with performance comparable to human experts.
• OASis: A humanness evaluation method that employs a 9-mer peptide search within the OAS database to produce granular, interpretable humanness scores that distinguish human and non-human sequences and correlate with clinical immunogenicity.
• Automated humanization: Automation of mouse-derived antibody sequence humanization using model-guided sequence generation.
• Natural repertoire leveraging: Utilization of the diversity of natural antibody repertoires from OAS to inform sequence design and humanness assessment.
• Sequence design for discovery: Computational generation and evaluation of candidate therapeutic antibody sequences for use in antibody discovery campaigns.

Scientific Applications:

• Antibody humanization: Humanizing mouse-derived antibody sequences for development of human-compatible therapeutics.
• Immunogenicity risk assessment: Assessing potential immunogenicity via humanness scoring and correlation with clinical immunogenicity data.
• Antibody discovery: Large-scale generation of human-like antibody sequences to expand candidate libraries in discovery campaigns.
• Benchmarking humanization strategies: Comparing computational humanization outputs to human expert benchmarks (177-antibody benchmark).

Methodology:

BioPhi uses a deep learning language model trained on the Observed Antibody Space (OAS) for sequence generation (Sapiens) and a 9-mer peptide search within OAS for humanness scoring (OASis), with benchmarking performed on 177 antibodies and evaluation of correlation to clinical immunogenicity.