Current advances in biopharmaceutical informatics: guidelines, impact and challenges in the computational developability assessment of antibody therapeutics

Rahul Khetan, Robin Curtis, Charlotte M. Deane, Johannes Thorling Hadsund, Uddipan Kar, Konrad Krawczyk, Daisuke Kuroda, Sarah A. Robinson, Pietro Sormanni, Kouhei Tsumoto, Jim Warwicker, Andrew C.r. MartinmAbs 2022 · DOI

TL;DR

This review highlights the growing importance of computational tools for assessing the "developability" of antibody-based therapeutics. By leveraging vast datasets and advanced bioinformatics, researchers can now predict and mitigate potential issues like aggregation, immunogenicity, and stability early in the drug discovery process, saving time and resources.

What Problem Does This Paper Address?

Developing new antibody-based drugs is a complex and expensive process. A significant challenge is ensuring that promising drug candidates are not only effective but also "developable" – meaning they can be manufactured reliably, are stable, and safe for patients. Many promising antibodies fail late in development due to unforeseen issues with their biophysical properties. This paper addresses the need for robust, early-stage methods to predict and address these developability challenges, moving beyond traditional experimental approaches.

Methods

The paper surveys a wide range of computational tools and approaches used in biopharmaceutical informatics for antibody developability assessment. Key methods include:

  • Homology Modeling and Structure Prediction: Predicting the 3D structure of antibodies from their amino acid sequences, crucial for understanding their physical properties.
  • Sequence-Based Analysis: Utilizing amino acid properties and patterns to predict stability, aggregation propensity, and immunogenicity.
  • Machine Learning and Deep Learning: Developing predictive models trained on large datasets of antibody sequences, structures, and experimental data to forecast various developability issues.
  • Database Curation and Data Mining: Integrating and analyzing data from diverse sources (patents, public databases, scientific literature) to build comprehensive datasets for model training and validation.
  • Developability Guidelines: Establishing criteria based on physicochemical properties of successful antibodies to guide the selection of new candidates.

Datasets

The paper references several key databases and datasets that are crucial for training and validating computational models:

  • Observed Antibody Space (OAS): Contains over a billion antibody sequences.
  • International Immunogenetics Information System (IMGT): Provides sequence, genome, and structure data.
  • Patented Antibody Database: Sequences from patent documents.
  • abysis: Integrated antibody sequence and structure management.
  • Protein Data Bank (PDB) & Structural Antibody Database (SAbDab): Collections of antibody structures.
  • Immune Epitope Database (IEDB): Experimental data on antibody and T-cell epitopes.
  • Jain dataset & TheraSabDab: Datasets of biophysical properties of approved mAbs.

The paper emphasizes the ongoing need for more curated and standardized datasets, particularly those linking experimental results with sequence and structural information.

Key Insights and Ideas

  • Developability is a Critical Early Hurdle: Ignoring developability early in the process leads to significant financial risk and wasted resources. Computational tools offer a way to de-risk development pipelines.
  • No Single Tool is a Silver Bullet: Developability assessment requires an orthogonal combination of different computational tools and metrics to mitigate method-specific biases.
  • Sequence is Powerful, but Structure Adds Depth: While sequence-based predictions are fast and scalable, incorporating structural information can provide more accurate insights into complex properties like aggregation.
  • "Human-likeness" Matters for Immunogenicity: Making antibodies as human-like as possible is a key strategy to reduce immunogenicity, and computational tools can help assess this.
  • Next-Generation Biotherapeutics Pose New Challenges: Engineered antibodies (e.g., bispecifics, ADCs) require specialized computational approaches due to their complex design and potential for unexpected liabilities.

Key Results

The paper doesn't present novel experimental results but rather synthesizes the current state of the art in computational developability assessment. It highlights:

  • The availability of numerous computational tools for predicting various developability issues, including solubility, aggregation, immunogenicity, and stability.
  • The successful application of these tools in screening antibody libraries and identifying candidates with favorable properties.
  • The development of "developability indices" and guidelines (e.g., Therapeutic Antibody Profiler - TAP) based on physicochemical properties of successful antibodies.
  • The increasing use of machine learning and deep learning to capture complex relationships between sequence, structure, and developability.

Limitations and Future Work

  • Data Scarcity and Heterogeneity: While improving, there's still a need for more standardized, high-quality, and publicly accessible datasets that link experimental outcomes with molecular data.
  • Impact of Formulation Conditions: Current computational tools often rely solely on sequence or structure and do not fully account for the influence of excipients and solution conditions on biophysical behavior. Future work should integrate these factors.
  • Industrial Validation: More information is needed on the validation of these computational tools in real-world industrial settings.
  • Next-Generation Biotherapeutics: Developing more accurate and precise computational methods is crucial for predicting the developability of complex engineered antibody formats.
  • AlphaFold's Potential: While AlphaFold shows promise for protein structure prediction, its efficacy in modeling antibodies and its integration into developability workflows remain to be proven.

Summary updated

Abstract

Therapeutic monoclonal antibodies and their derivatives are key components of clinical pipelines in the global biopharmaceutical industry. The availability of large datasets of antibody sequences, structures, and biophysical properties is increasingly enabling the development of predictive models and computational tools for the "developability assessment" of antibody drug candidates. Here, we provide an overview of the antibody informatics tools applicable to the prediction of developability issues such as stability, aggregation, immunogenicity, and chemical degradation. We further evaluate the opportunities and challenges of using biopharmaceutical informatics for drug discovery and optimization. Finally, we discuss the potential of developability guidelines based on <i>in silico</i> metrics that can be used for the assessment of antibody stability and manufacturability.
BibTeX
@inproceedings{khetan2022current,
  author = {Rahul Khetan and Robin Curtis and Charlotte M. Deane and Johannes Thorling Hadsund and Uddipan Kar and Konrad Krawczyk and Daisuke Kuroda and Sarah A. Robinson and Pietro Sormanni and Kouhei Tsumoto and Jim Warwicker and Andrew C.R. Martin},
  title = {{Current advances in biopharmaceutical informatics: guidelines, impact and challenges in the computational developability assessment of antibody therapeutics}},
  booktitle = {mAbs},
  year = {2022},
  doi = {10.1080/19420862.2021.2020082},
}
Figure