Automated Machine Learning Platform for Discovering and Optimizing Protein-Stabilizing Synthetic Polymers

Background

The field of therapeutic protein stabilization is critical for advancing treatments for a variety of diseases, particularly those involving enzymes with potential for tissue regeneration. In the context of central nervous system (CNS) injuries, enzymes like Chondroitinase ABC (ChABC) are of significant interest due to their ability to degrade inhibitory molecules that impede neural repair. However, the clinical translation of such enzymes is severely hampered by their inherent instability under physiological conditions, leading to rapid loss of activity and, consequently, limited therapeutic efficacy. The need for technologies that can maintain enzyme activity over extended periods is especially pressing in applications like neural regeneration, where sustained enzymatic action is required to achieve meaningful outcomes.

Current approaches to protein stabilization, such as the use of sugar-based excipients, immobilization techniques, or protein engineering, have not been able to adequately address the instability of enzymes like ChABC. These methods often provide only modest improvements in stability and can introduce new challenges, such as reduced enzyme activity, increased immunogenicity, or difficulties in large-scale production. Furthermore, the traditional process of discovering suitable stabilizing agents is labor-intensive, slow, and largely reliant on trial-and-error experimentation, making it inefficient for systematically optimizing complex protein-polymer interactions. As a result, there remains a significant unmet need for robust, scalable, and data-driven strategies that can rapidly identify and optimize stabilizing formulations tailored to the unique requirements of therapeutic enzymes.

Technology

This technology is an automated, machine learning-driven platform designed to discover and optimize synthetic copolymers that stabilize therapeutic proteins, with a primary focus on the enzyme Chondroitinase ABC (ChABC) for neural regeneration. The system integrates a robotic liquid handling platform to perform high-throughput PET-RAFT polymerization, generating extensive libraries of copolymers with diverse monomer compositions and chain lengths. These copolymers are systematically complexed with ChABC and screened for their ability to retain enzymatic activity under physiological and denaturing conditions. Experimental data—including polymer composition, molecular weight, and hydrophobicity—are analyzed using advanced machine learning models such as Gaussian Process Regression and random forest algorithms. An active learning loop iteratively proposes, synthesizes, and tests new copolymer candidates, continually refining the model to maximize enzyme stability and performance. The resulting polymer-enzyme complexes demonstrate enhanced catalytic activity, long-term stability, and biocompatibility, with no observed cytotoxic or inflammatory effects in biological assays.

What differentiates this technology is its seamless integration of automation, high-throughput experimentation, and intelligent data-driven optimization. Unlike traditional protein stabilization methods that rely on manual, trial-and-error approaches or limited chemical modifications, this platform leverages robotic synthesis and machine learning to rapidly explore vast chemical spaces and identify optimal polymer-protein interfaces. The active learning framework ensures that each experimental cycle is informed by previous results, accelerating the discovery of high-performing copolymers with minimal human intervention. Additionally, the platform’s modular design allows it to be generalized for stabilizing a wide range of enzymes and proteins, not just ChABC, making it a versatile tool for biopharmaceutical development and industrial enzyme formulation. This combination of scalability, adaptability, and predictive power positions the technology as a transformative solution for overcoming the stability challenges that limit the therapeutic and industrial use of proteins.

Benefits

• Enables automated, high-throughput synthesis and screening of synthetic copolymers for protein stabilization.
• Uses machine learning-driven active learning to efficiently optimize copolymer compositions for enhanced enzyme stability and activity.
• Produces polymer-enzyme complexes that maintain or improve enzyme catalytic activity at physiological temperatures over extended periods.
• Demonstrates improved long-term thermal stability of enzymes compared to conventional stabilizers.
• Ensures biocompatibility by showing no cytotoxicity or inflammatory response in relevant cell cultures.
• Offers a generalizable platform applicable to diverse enzymes and proteins beyond Chondroitinase ABC.
• Accelerates discovery and optimization of stable biopharmaceutical formulations, reducing reliance on trial-and-error methods.

Applications

• Stabilizing therapeutic enzymes for injection
• Developing long-acting protein drugs
• Enhancing industrial enzyme formulations
• Automated polymer discovery platforms

Additional info

This automated platform integrates robotic liquid handling and PET-RAFT polymerization to synthesize diverse copolymer libraries. An active learning loop employs Gaussian Process Regression and random forest models to optimize monomer compositions and chain lengths. These synthetic polymers form complexes with enzymes like Chondroitinase ABC, significantly enhancing thermal stability and catalytic activity under physiological conditions for neural regeneration while ensuring biological safety and non-toxicity.