For decades, biologics discovery has balanced careful experimentation with the realities of cost, risk, and time. Scientists have developed a sophisticated toolkit for discovering monoclonal antibodies and other therapeutic proteins; however, these tools are relatively costly and labor-intensive. The question ‘Can we design better biologics from the ground up with speed and specificity?‘
A new wave is transforming that landscape. AI and machine learning are more than hype. It is the convergence of biotech and digital transformation, with direct impact on how quickly, safely, and cost-effectively new medicines reach patients.
The Traditional Foundations of Biologics Discovery
Historically, biologics discovery relied on biological systems to generate and refine candidates. Two workhorse strategies remain central.
In hybridoma technology, animals (often mice) are immunized with a target antigen, and their immune system produces a diverse array of antibodies. Although it allows the isolation of monoclonal antibodies with high affinity, this technology faces challenges such as species differences and ethical considerations [1, 2].
Phage and yeast display methods involve presenting antibody fragments on the surfaces of phages or yeast cells. Phages or yeasts binding to the target antigen are further amplified. It enables the rapid screening of large antibody fragment libraries, allowing for the selection of variants with the best binding properties. While high-throughput and versatile, these methods often require extra steps to optimize antibody stability, manufacturability, and developability [1, 2].
These approaches can be slow, labor-intensive, and costly, especially in the early stages, where many candidates fail due to poor biophysical properties or immunogenicity.
De-Risking Early Development with In Silico Testing
Late-stage failures are expensive, but can be avoided. That is where in silico testing comes to the rescue. Using computational tools early in the pipeline, teams can screen vast libraries for key liabilities, such as developability, immunogenicity, and manufacturability.
Algorithms predict solubility, aggregation, and stability metrics directly from sequence or structural data [1, 3-6]. Tools like CamSol and QSPR models help filter out problematic candidates early.
Platforms like Ardigen’s ARDisplay I & II and ARDiTox predict potential T-cell epitopes and assess the risk of off-target immune responses, respectively, reducing the risk of adverse immune reactions in patients [7, 8].
In silico assessments evaluate expression profiles, biophysical properties, and similarity to successful marketed biologics, guiding selection toward candidates with potential for production scale-up [1, 3].
These computational strategies reduce the need for costly in vitro assays, streamline experimental focus, and increase the likelihood of clinical and commercial success. Organizations that adopt predictive models early move away from a failure-prone trial-and-error process toward rational and data-driven decisions.
Transforming Structure Prediction with AI
Protein structure prediction was once a bottleneck in biologics design. Now AI helps surpass that barrier. Deep learning models, such as AlphaFold, and generative approaches, like RFDiffusion, have set new standards.
AlphaFold predicts 3D protein structures from amino acid sequences with near-experimental accuracy. Including those proteins without known homologs, by predicting inter-residue distances and orientations using deep neural networks [9-15]. This dramatically reduces the time and cost of understanding target or therapeutic protein conformations.
RFDiffusion uses generative models to create entirely new protein structures. It extends beyond predicting existing proteins to allow the design of novel ones with tailored properties [9, 15].
Ardigen’s Prism offers a wide range of protein LLMs that complement previous models and support precise protein engineering.
AI Method | Core Approach & Impact | Notable Examples |
|---|---|---|
Deep Learning | Uses neural networks to learn sequence-structure relationships from large datasets. | AlphaFold, RoseTTAFold, Boltz-1, Boltz-2 and Chai-1 |
Generative Models | Designs new protein structures by generating plausible 3D conformations from scratch. | RFDiffusion, Chroma, AbDesign & AbDock, DiffAb, Guided DiffAb, EAGLE |
Language Models | Leverage evolutionary patterns in protein sequences to predict structures at scale and enhance protein engineering. | ESMFold, Prism |
Tab. 1. Examples of AI-driven solutions for protein structure discovery and design.
This shift from experimental-only to AI-augmented prediction closes gaps in template-free modeling and accelerates hit-to-lead workflows. It also enhances integration with experimental methods, such as cryo-EM, helping researchers interpret mutational effects or design variants with greater confidence [6, 16].
Yet challenges remain. Multi-domain proteins, protein complexes, membrane proteins, intrinsically disordered proteins, dynamic conformational changes, and the effects of post-translational modifications continue to challenge current AI models [6, 12, 15]. Nevertheless, the leap in capability has fundamentally changed how discovery teams think about feasibility, timelines, and risk.
Machine Learning for Optimization: Stability, Binding, Manufacturability
AI does not stop at structure prediction. Machine learning is now a critical partner in optimizing lead candidates. It can utilize sequence and structure data of antibodies and their targets to maximize core traits.
Machine learning models can reduce the time spent on stability experiments by identifying protein sequences with enhanced thermal stability and reduced aggregation propensity [17].
Deep learning also plays a significant role in balancing affinity and specificity. It can suggest targeted mutations that improve binding to the desired target while minimizing off-target effects [18-24]. These models even predict binding constants based on sequence or structural data, making it easier to prioritize and engineer leads.
Additionally, machine learning facilitates manufacturability assessments by enabling teams to select candidates with desirable properties, such as low viscosity, high solubility, high expression yields, and low self-association [17, 22]. These traits are essential for scalable, cost-effective production.
As companies seek to co-optimize multiple properties simultaneously, AI reveals trade-offs that would be hard to spot experimentally.
The Future: A Digitally Transformed Discovery Pipeline
Biotech and pharma companies that integrate AI across biologics discovery, engineering, and development may count on:
- Faster timelines from target to clinic.
- Higher success rates through early de-risking.
- Better optimized, more manufacturable products.
- More personalized and precise therapies.
However, adoption comes with challenges. Regulatory agencies will need confidence in AI models, especially given their “black box” reputation. Interpretability and explainability will be crucial, not just for approval, but for internal scientific rigor.
Companies do not need to build this AI capability from scratch. At Ardigen, we partner with biopharma innovators to integrate advanced AI solutions across the entire discovery pipeline – from identifying new biologics, through immunogenicity, stability, and affinity assessments, to predicting manufacturability. We combine deep domain expertise with machine learning and offer solutions that help teams accelerate candidate selection, reduce risk, and deliver better biologics, faster.
Author: Martyna Piotrowska |
Technical editing: Ardigen expert Joanna Marczyńska-Grzelak
Bibliography:
- Fernández-Quintero M, et al. Assessing developability early in the discovery process for novel biologics. mAbs. 2023;15. https://doi.org/10.1080/19420862.2023.2171248
- McGonigle P. How biologics have changed the drug discovery landscape. Annu Rev Pharmacol Toxicol. 2024. https://doi.org/10.1146/annurev-pharmtox-061724-080811
- Ahmed L, et al. Intrinsic physicochemical profile of marketed antibody-based biotherapeutics. Proc Natl Acad Sci U S A. 2021;118. https://doi.org/10.1073/pnas.2020577118
- Pérez A, et al. In vitro and in silico assessment of the developability of a designed monoclonal antibody library. mAbs. 2019;11:388–400. https://doi.org/10.1080/19420862.2018.1556082
- Sankar K, et al. A descriptor set for QSPR prediction in biologics. Mol Inform. 2022;41. https://doi.org/10.1002/minf.202100240
- Versini R, et al. A perspective on the prospective use of AI in protein structure prediction. J Chem Inf Model. 2023. https://doi.org/10.1021/acs.jcim.3c01361
- Mecklenbräuker S, et al. Identification of tumor-specific MHC ligands through improved biochemical isolation and incorporation of machine learning. bioRxiv. 2023. https://doi.org/10.1101/2023.06.08.544182
- Murcia Pienkowski V, et al. ARDitox: Platform for the prediction of TCRs potential off-target binding. bioRxiv. 2023. https://doi.org/10.1101/2023.04.11.536336
- Chen L, et al. AI-driven deep learning techniques in protein structure prediction. Int J Mol Sci. 2024;25. https://doi.org/10.3390/ijms25158426
- Pakhrin S, et al. Deep learning-based advances in protein structure prediction. Int J Mol Sci. 2021;22. https://doi.org/10.3390/ijms22115553
- Pearce R, Zhang Y. Toward the solution of the protein structure prediction problem. J Biol Chem. 2021;297. https://doi.org/10.1016/j.jbc.2021.100870
- Peng C, et al. Recent advances and challenges in protein structure prediction. J Chem Inf Model. 2023. https://doi.org/10.1021/acs.jcim.3c01324
- Qiu X, et al. Advances in AI for protein structure prediction: Implications for cancer drug discovery and development. Biomolecules. 2024;14. https://doi.org/10.3390/biom14030339
- Senior A, et al. Improved protein structure prediction using potentials from deep learning. Nature. 2020;577:706–710. https://doi.org/10.1038/s41586-019-1923-7
- Wuyun Q, et al. Recent progress of protein tertiary structure prediction. Molecules. 2024;29. https://doi.org/10.3390/molecules29040832
- Ohno S, et al. Prediction of protein structure and AI. J Hum Genet. 2024:1–4. https://doi.org/10.1038/s10038-023-01215-4
- Narayanan H, et al. Design of biopharmaceutical formulations accelerated by machine learning. Mol Pharm. 2021. https://doi.org/10.1021/acs.molpharmaceut.1c00469
- Feng H, et al. Machine learning-driven methods for nanobody affinity prediction. ACS Omega. 2024;9:47893–47902. https://doi.org/10.1021/acsomega.4c09718
- Guo Z, Yamaguchi R. Machine learning methods for protein–protein binding affinity prediction in protein design. Front Bioinform. 2022;2. https://doi.org/10.3389/fbinf.2022.1065703
- Li L, et al. Machine learning optimization of candidate antibody yields highly diverse sub-nanomolar affinity antibody libraries. Nat Commun. 2022;14. https://doi.org/10.1101/2022.10.07.502662
- Makowski E, et al. Co-optimization of therapeutic antibody affinity and specificity using machine learning models that generalize to novel mutational space. Nat Commun. 2022;13. https://doi.org/10.1038/s41467-022-31457-3
- Makowski E, et al. Optimization of therapeutic antibodies for reduced self-association and non-specific binding via interpretable machine learning. Nat Biomed Eng. 2023:1–12. https://doi.org/10.1038/s41551-023-01074-6
- Rube H, et al. Prediction of protein–ligand binding affinity from sequencing data with interpretable machine learning. Nat Biotechnol. 2022;40:1520–1527. https://doi.org/10.1038/s41587-022-01307-0
- Zheng F, et al. Systematic investigation of machine learning on limited data: A study on predicting protein–protein binding strength. Comput Struct Biotechnol J. 2023; 23:460–472. https://doi.org/10.1016/j.csbj.2023.12.018
- Lin Z, et al. Evolutionary-scale prediction of atomic level protein structure with a language model. bioRxiv. 2022. https://doi.org/10.1126/science.ade2574
- Salman M, et al. Harnessing artificial intelligence for drug design. Acta Sci Microbiol. 2024. https://doi.org/10.31080/asmi.2024.07.1390
- Schauperl M, Denny R. AI-based protein structure prediction in drug discovery. J Chem Inf Model. 2022. https://doi.org/10.1021/acs.jcim.2c00026
- Willis L, et al. Biophysical analysis of therapeutic antibodies. Biologics. 2024;18:413–432. https://doi.org/10.2147/BTT.S486345
