What can AI-designed proteins actually do right now?
AI-driven protein design has already moved beyond theory into practical, high-impact applications. During the COVID-19 pandemic, AI-directed computational design helped accelerate antigen characterization and the development of broadly neutralizing antibodies, acting as a real-world stress test for the technology [2]. In a more recent example, researchers used a machine learning-guided reverse vaccinology approach to scan the entire proteome of Staphylococcus aureus and identified a novel vaccine candidate (IMTS8) that, when tested in mice, elicited strong antibody responses and provided near-complete protection against a multi-drug-resistant strain [7]. This shows that AI can successfully pinpoint effective therapeutic candidates from scratch, at least in animal models.
Beyond vaccines, AI is being used to design entirely new proteins that have never existed in nature. A 2023 review highlights how protein engineers are combining machine learning tools with structure-predicting models like AlphaFold2 to pursue sophisticated de novo designs [6]. Startups like Generate Biomedicines have raised substantial funding ($370 million) with the goal of creating therapeutic proteins—including antibodies, cytokines, and enzymes—from scratch for any disease, and they planned to advance multiple programs into clinical trials in 2023 [8]. These developments indicate that the design pipeline is operational and producing candidates ready for human testing.
What is the main barrier to clinical deployment?
The single biggest obstacle is what researchers call the 'validation gap'—the disconnect between AI-predicted protein structures and the experimental proof required for clinical use. A critical review notes that while AI models like AlphaFold can predict structures with high accuracy, there are currently no clear guidelines for qualifying AI-derived structural data for therapeutic development, and case studies show that predictions can fail when not backed by wet-lab experiments [3]. This gap means that even the most promising AI-designed proteins must still undergo the same lengthy, expensive experimental validation as traditionally discovered drugs.
The COVID-19 pandemic exposed additional limitations. Despite AI's contributions, challenges related to data bias, model interpretability, and the bottleneck of experimental validation became apparent [2]. The rapid pace of AI innovation often outstripped established regulatory pathways, raising questions about standardization, validation, and long-term safety [2]. A 2025 review on antibody design further notes that many computational methods are limited by the unique structural biology of antibodies, meaning tools that work well on generic proteins may not directly apply to this key therapeutic class [4]. These issues collectively mean that while AI can accelerate discovery, it cannot yet bypass the rigorous clinical testing and regulatory approval process.
Do the studies agree on the readiness of AI-designed proteins?
The studies broadly agree on two points: AI has transformative potential for protein design, and significant translational hurdles remain. Papers [1], [2], [5], and [6] all emphasize the power of AI to accelerate design and discovery, while [2], [3], and [4] converge on the need for better experimental validation and regulatory frameworks. There is no fundamental disagreement—rather, the papers address different aspects of the pipeline. For instance, [7] provides a concrete success story in a mouse model, while [3] focuses on the structural validation gap, and [2] highlights regulatory and data-bias issues. Together, they paint a consistent picture: the design phase is advancing rapidly, but the translation to clinical use is lagging.
One area where the evidence is thinner is in human clinical trials. None of the provided abstracts report results from a completed human trial of an AI-designed protein. The closest is the planned 2023 trials by Generate Biomedicines [8], but no outcomes are reported here. This absence underscores the central answer: AI-designed proteins are not yet clinically deployed, but the foundational work—from de novo design [6] to preclinical protection in animals [7]—is laying the groundwork for that next step.
About These Sources
This answer is built on 8 peer-reviewed studies — published from 2021 to 2026, 5 from 2024 or later, 5 in Q1 journals — selected as the most relevant from 9 studies that passed quality screening, drawn from 66 papers retrieved from a database of over 500 million.
Sources used in this answer
Artificial intelligence driven protein design and sustainable nanomedicine for advanced theranostics
Reviews how AI-driven protein design, combined with sustainable nanocarriers, is enabling precise theranostics (diagnosis + therapy) for oncology and other diseases, but notes translational challenges remain.
Artificial intelligence directed computational protein design: lessons from COVID-19 for pandemic-ready vaccines and antibody therapeutics
Using COVID-19 as a case study, this review finds that AI-directed protein design accelerated vaccine and antibody development but revealed critical limitations in data bias, model interpretability, experimental validation, and regulatory integration.
From AlphaFold to the Clinic: A Critical Review of the Validation Gap in AI-Predicted Protein Structures for Therapeutic Development
This critical review identifies a 'validation gap' between AI-predicted protein structures and the experimental confirmation needed for clinical use, noting the absence of clear guidelines for qualifying AI-derived structural data.
Applying computational protein design to therapeutic antibody discovery - current state and perspectives
Reviews computational methods for antibody design and finds that many tools are limited by the unique structural biology of antibodies, requiring specialized approaches for binder discovery.
From <i>De Novo</i> Design to Redesign: Harnessing Computational Protein Design for Understanding SARS-CoV-2 Molecular Mechanisms and Developing Therapeutics
Describes how computational protein design, using tools like AlphaFold and RFdiffusion, has been used to develop peptide inhibitors, nanobodies, and monoclonal antibodies against SARS-CoV-2, pending regulatory approval.
AI-enhanced protein design makes proteins that have never existed
Reports that protein engineers are using evolving machine learning tools and AlphaFold2 to pursue more sophisticated de novo protein designs that have never existed in nature.
A novel hypothetical protein (SAUSA300_1684) confers excellent protection against multi-drug-resistant Staphylococcus aureus infection in the murine model
Using a machine learning-guided reverse vaccinology approach, this study identified a novel vaccine candidate (IMTS8) that conferred near-complete protection against multi-drug-resistant Staphylococcus aureus in a murine infection model.
Generate raises funds for de novo proteins
Reports that Generate Biomedicines raised $370 million to develop de novo protein design technology, with plans to advance multiple therapeutic programs into clinical trials in 2023.