Computational counterselection identifies nonspecific therapeutic biologic candidates

Area of Science:

• Biotechnology
• Immunology
• Computational Biology

Background:

• Effective biologics demand high specificity and minimal off-target binding, properties not guaranteed by current discovery methods.
• Molecular counterselection is an experimental technique to identify and remove nonspecific sequences, but it is costly and often inefficient.

Purpose of the Study:

• To introduce computational counterselection, a machine learning framework for identifying specific biologics.
• To demonstrate the efficacy of computational counterselection compared to molecular counterselection in antibody discovery.

Main Methods:

• Developed and applied machine learning models for computational counterselection using sequencing data from single-target affinity selection of antibodies.
• Validated the method by comparing its performance against molecular counterselection using cross-target selection and individual binding assays.

Main Results:

• Computational counterselection effectively identified and eliminated off-target, nonspecific antibodies while retaining on-target, specific antibodies.
• The computational approach outperformed molecular counterselection in specificity and efficiency.
• A general model identified generally polyspecific antibody sequences based on affinity data from diverse targets.

Conclusions:

• Computational counterselection offers a more efficient and effective alternative to molecular counterselection for discovering specific biologics.
• This AI-driven framework significantly improves the identification and elimination of nonspecific antibody sequences.
• The method facilitates the discovery of highly specific antibodies and the characterization of polyspecific sequences.