Multidimensional maturation of antibody variable domains with machine-learning assistance

Area of Science:

• Biotechnology
• Immunology
• Computational Biology

Background:

• Traditional antibody development using mutagenesis can negatively impact biophysical properties.
• Camelid heavy-chain antibody variable domains (VHHs) are crucial for therapeutic antibody development.

Purpose of the Study:

• To simultaneously enhance antibody affinity and expression levels of VHH domains.
• To apply machine learning-assisted molecular evolution for multidimensional antibody optimization.

Main Methods:

• Phage display and deep sequencing were used for affinity maturation of an anti-SARS-CoV-2 VHH.
• Machine learning models were trained on experimentally measured expression levels and affinities of 117 variants.
• High-ranking variants were predicted and validated computationally.

Main Results:

• Selected VHH variants demonstrated 50-70 fold increased affinity in the picomolar range.
• Engineered VHHs exhibited 4-5 fold higher expression levels compared to wild-type.
• One variant displayed a 9.5°C improvement in thermal stability.

Conclusions:

• Machine learning-assisted molecular evolution is effective for multidimensional antibody property optimization.
• This strategy successfully improved both affinity and expression levels of VHH domains.
• The developed VHH variants show promise for therapeutic applications.