Predictive stability modeling is revolutionizing biologic drug shelf-life evaluation by providing accurate long-term stability forecasts based on relevant short-term data. Several practical applications will be presented. The integration of scientific rigor and statistical robustness in these models supports critical CMC decision-making, optimizes stability filing strategies, and accelerates the development timeline for new biologic therapies.

High-concentration monoclonal antibody (mAb) formulations are limited by viscosity driven by protein–protein interactions, yet early risk assessment lacks high-throughput tools. We developed a high-throughput SAXS workflow to detect self-association at dilute concentrations. Across 22 mAbs in histidine buffer (pH 6.0), low-q structure factor transitions below 10 mg/mL predicted viscosity at 150 mg/mL, correctly classifying 21/22 antibodies. This scalable, sample-efficient approach enables early developability screening and model validation.

Ensuring consistent enzyme performance is crucial in regulated drug process development. Traditional optimization methods, such as one-factor-at-a-time (OFAT) and design of experiments (DoE), can be resource-intensive and biased. Here, we used Bayesian Optimization (BO) to rapidly optimize a ligase enzyme activity assay used in process development. BO rapidly identified settings that matched or exceeded OFAT-derived performance while reducing experimental workload by approximately 85% versus OFAT and 65% versus DoE. The resulting model also produced accurate robustness predictions that were validated experimentally. Beyond performance gains, BO delivered interpretable parameter insights, informing mechanistic understanding and enabling more efficient method transfer.

Microfluidic electrophoresis is a powerful characterization technique for both novel protein-based therapeutics and protein biomarkers. We will discuss innovative methods that use both size-based and charge-based automated microfluidic electrophoresis to analyze different types of proteins and how this translates to the drug discovery and development process. We will also discuss how our predictive analysis method can be used for reagent manufacturing protocols for microfluidics applications.

Developability of biotherapeutics usually involves several stages of preclinical experiments to inform the manufacturability, long-term stability, formulation, and delivery of a biotherapeutic. Over the past two years, data capture systems have been deployed at AbbVie for pipeline use, with the benefit of enabling collection of AI/ML ready datasets. Using these tools and high-throughput screening assays, we are collecting developability data on thousands of IgG1 mAbs to enable correlative and predictive modeling. This talk will cover the data capture systems that enable such an initiative, as well as current experimental and model status.

High-concentration therapeutic antibody development is constrained by viscosity and limited experimental throughput, motivating in silico screening. A multimodal feature learning (MMF) approach combines conventional MOE protein descriptors with protein large-language-model embeddings to improve viscosity prediction while enabling feature interpretability. Across regression and classification benchmarks (including internal data), MMF shows improved performance versus prior models and supports automated high-throughput candidate triage.

This presentation highlights how mechanistic modeling and machine learning can enhance analytical technologies in recombinant adeno-associated virus (rAAV) manufacturing. Case studies showcase models built from diverse analytical datasets that accelerate analytical development and translate analytical insights into effective process control. Additionally, a novel real-time titer quantification approach leveraging single-cell measurements and machine learning is demonstrated across a range of rAAV constructs.

Viruses undergo rapid genetic changes that affect their virulence, a phenomenon that was evident during the COVID pandemic with the emergence of variants that could evade protection offered by some vaccines. The SARS-CoV-2 variants were mapped by mutations in the amino acid sequence of the spike protein of the virus. Changes to the spike protein variants could also be mapped by glycan profiling using robust and sensitive methods of analysis. Such changes affected the interaction of the virus to the ACE-2 receptor that facilitated cellular infection.

Digital transformation, the accelerating pace of scientific and technological advancement and increasingly sophisticated analytics and computational methods are putting intense pressure on delivering digital solutions in biotherapeutics research and development. We'll discuss AbbVie's work in this space over the last decade to share lessons learned, discuss challenges and opportunities, and share our vision of a digitally enabled and data-driven biotherapeutics organization to meet the challenges of tomorrow's biopharma landscape.

In biopharmaceutical development, accurately tracking cell-line selection and sample lineage is crucial for maintaining product quality and meeting regulatory requirements. Implementing a digital system that records cell line genealogy and connects analytical results to stage-specific samples from cell-line screening ensures traceability and upholds data integrity. This approach aligns with FAIR data principles and facilitates informed decision-making throughout cell-line development and analysis.

Monoclonal antibody heterogeneities, including charge variants and aggregates, can influence immunogenicity through altered cytokine responses. We combined human PBMC-based cytokine release assays with mechanistic modeling and deep neural network regression to evaluate immunogenic risk. Trastuzumab charge variants and stress-induced aggregates of rituximab and infliximab displayed distinct cytokine profiles linked to functional differences. The integrated experimental–computational framework successfully predicted cytokine dynamics, offering a translational approach to assess and mitigate immunogenicity risks during biotherapeutic development.

Cells grown in perfusion culture are assumed to be in a pseudo steady state; however, changes in productivity and product quality are often observed, and the contributing factors are not yet identified. Here, we use single-cell RNA sequencing to show how the evolution of heterogeneous cell populations in perfusion culture contribute to inconsistencies in monoclonal antibody production. These data are used to drive process improvements for improved robustness.

In silico computations play an increasing role in drug development, but platforms combining multiple models and comprehensively evaluating therapeutic proteins' developability are currently lacking. This presentation covers this gap, showing how different models, spanning structure prediction, bioinformatics, machine learning, and molecular dynamics, can be combined within an automated platform to speed up and de-risk candidate selection, lead characterization, and formulation development. Relevant case studies will be presented.