A new machine learning framework is transforming the early evaluation of biotherapeutics, particularly monoclonal antibodies, allowing researchers to assess and optimize drug-target interactions more efficiently. The framework utilizes minimal physiologically based pharmacokinetic (mPBPK) modeling coupled with machine learning to help researchers infer optimal antibody properties and their targeted engagements.
The development of therapeutic antibodies has long been constrained by the limited knowledge of both the antibodies and their target pharmacology during initial optimization stages. Decisions made at this early phase can critically determine the success or failure of drug candidates. To mitigate these challenges, researchers at Sanofi have introduced this innovative machine learning-based target pharmacology assessment framework.
Traditionally, the optimization of antibody candidates occurs within time-consuming and resource-intensive processes, requiring comprehensive pharmacological studies for each candidate. This new approach promises to speed up the early evaluation process dramatically.
By employing the mPBPK model, the researchers simulated interactions between numerous virtual antibody candidates—varying both their physicochemical properties (such as charge and binding affinity) and target characteristics (baseline expression and half-life). Their study involved high-throughput virtual screenings of drug-target pairs, analyzing how variations among candidates impact target occupancy (TO), defined as the ratio of occupied targets to available targets.
The findings reveal significant relationships between variables, demonstrating dependencies of TO on factors like antibody dose and target type. For example, the study identifies specific conditions under which maximum target engagement is likely to occur, such as maintaining low binding constants for certain target baselines.
"By unraveling new design rules for antibody drug properties, we deliver a first-in-class ML-based target pharmacology assessment framework toward achieving favorable drug-target interactions," explained the researchers. This framework extends the capabilities of existing modeling approaches by integrating insightful machine learning analyses, resulting in improved decision-making early on.
The machine learning classifier not only helps to identify optimal drug-target combinations but also categorizes candidates based on their predicted TO percentages. Candidates achieving over 90% TO are likely to meet necessary pharmacological criteria, enabling researchers to focus on the most promising options.
The results demonstrate this methodology's potential for enhancing lead discovery and optimizing drug development pipelines. With current iterations successfully narrowing the initial pool of 11,620 candidates down to approximately 590 optimal candidates based on desired TO levels, this mPBPK framework is poised to revolutionize early-stage drug development.
While the model is not without limitations, researchers state it is valuable as an early investigation tool for assessing drug-target properties. Future updates could incorporate additional data from clinical studies and complementary pharmacokinetic parameters to deepen the framework's predictive capabilities.
Advancements propelled by this new framework could fundamentally reshape strategies for therapeutic antibody discovery and development, accelerating progress toward viable drug candidates and improving overall efficiencies within the biopharmaceutical pipeline.