Abstracts

AI for Biotechnology: Technological Advancements, Readiness and Policy Implications

Nick Vangheluwe (European Commission)

AI is rapidly expanding across the biotechnology lifecycle in the health, industrial, and agri-food sectors, playing an instrumental role in translating scientific discoveries into large-scale development, manufacturing, and market deployment. The European Commission has prioritised the safe and responsible adoption of AI in Science, including biotechnology.

Building on the recent in-depth study by the Joint Research Centre on the capabilities and trends of AI models applied to biology (doi:10.2760/9397036), this presentation will provide a state-of-the-art overview of recent advancements in machine learning and generative AI in the field and clarify the opportunities, challenges and readiness.

Furthermore, it will explore the policy implications of these developments in the context of the European Biotech Act proposed on 16 December 2025. This policy initiative aims to facilitate AI integration across the EU biotechnology ecosystem and accelerate innovations, products and services for the Union’s single market while ensuring the highest safety standards

De novo protein sequencing with deep learning: discovery, therapeutics and biosecurity

Konstantinos Kalogeropoulos (Technical University of Denmark)

Proteins carry out most functions in the cell, and proteomics is the large-scale study of the proteins present in a biological sample. The standard method for proteomics is mass spectrometry, which measures peptides derived from those proteins and identifies their sequences by detecting their fragments. While genomes indicate what an organism can produce, proteomics measures what is actually present.

Conventional proteomics analysis does not read sequences directly. Measured fragment patterns are matched against sequences predicted from a reference genome, so only proteins already represented in the database can be identified. This limits the analysis of protein therapeutics such as antibodies, which are not encoded in any reference genome, of organisms without a sequenced genome, and of engineered or computationally designed proteins.

De novo peptide sequencing removes this dependency by inferring the amino acid sequence directly from the fragmentation spectrum. Accuracy was long insufficient for routine use, but deep learning models have made the approach practical. I will outline how these models work and what they enable, such identification of proteins and organisms absent from reference databases, sequencing of therapeutics, and detection of novel or engineered proteins, which is relevant to biosafety and biosecurity. I will conclude with the current limitations of these methods and their implications.

A Smarter Lab? Ethical Implications of AI Use on Biomedical Expertise

Isis Hazewindus (Leiden University Medical Center)

Bringing AI into the laboratory is being presented as a solution to myriad problems. From helping researchers limiting the search space of a problem, to the possibility of automating the entire Design-Build-Test-Learn cycle or the development of entirely new biological agents - there is an AI system for any scientific aim. But what does the use of AI systems mean for biomedical expertise, and what are its ethical implications? This talk will explore the possible consequences of AI use for scientific values, epistemic practices and expertise in the lab.

Generative AI-guided design of bacteriophages

Samuel Hyo-nam King (Stanford University) 

All the innovations of the natural world are written in genomes. Harnessing their capabilities could transform technologies across medicine, materials, and sustainability. However, even the simplest genomes are extraordinarily complex. Genome language models are a new class of artificial intelligence (AI) algorithms that work much like typical LLMs, except instead of being trained on text from the internet, they're trained on millions of genomes spanning all of life. This gives them fluency in the language of DNA and the evolutionary constraints that shape genomes in nature. In this talk, I will discuss our efforts using genome language models to generate functional viruses that infect bacteria (bacteriophages), and their potential as therapeutics for antibiotic resistance. I will also highlight the biosecurity aspects of this research and our work on improving provenance of AI-generated genomes.

AI Biotechnology convergence in the strategic age: a double governance challenge

Ruth Mampuys (The Netherlands Scientific Council for Government Policy)

 The convergence of artificial intelligence (AI) and biotechnology is transforming how biological data is produced, circulated and governed. AI-driven tools increasingly rely on and generate digital sequence data, shifting biological data from an input in biotechnological research to a starting point for innovation. This transformation raises questions concerning biosafety, biosecurity, privacy and data governance, while challenging the adequacy of existing regulatory frameworks.

In addition, these developments unfold within a context of intensifying geopolitical rivalry, in which biotechnology is increasingly framed as a strategic and security asset. The resulting pressures to accelerate innovation, strengthen national competitiveness and safeguard security introduce additional tensions for scientists and policymakers, particularly concerning international scientific cooperation, academic freedom and the long-term societal consequences of technological development.

Drawing on insights from two complementary analyses, this presentation develops the notion of a double governance challenge: governing the novel risks and ethical questions arising from AI–biotechnology convergence, while simultaneously navigating the geopolitical dynamics and competing rationalities shaping its development. Governing AI biotechnology convergence requires attention not only to the capabilities and risks of emerging technologies, but also to the political and institutional conditions in which they are developed and deployed.

AI and Biotech: from data to technological innovation

Sanne Abeln

Artificial intelligence (AI) is developing at a rapid pace, offering promising opportunities for biotechnology. Important recent breakthroughs in AI include deep learning, generative AI, and multimodal AI. Deep learning enables highly accurate predictions based on large datasets. Generative AI can create new designs, such as images or molecular structures. Multimodal AI combines different data sources to make complex predictions.

In biotechnology, AI can now make highly accurate predictions by learning from large amounts of data. In the future, AI will also be able to learn from limited datasets, supported by large public biological databases. The consequences for biotechnology are significant. AI makes it possible to create new biological designs for new drugs, crops, biofuels, and antibodies. The biotechnological innovation process can also become much faster and more efficient through robotization and smart AI-supported measurement methods. Additionally, AI can support complex decisions by integrating highly diverse data sources.

However, there are also challenges: fragmented data can slow down progress. Furthermore, AI models themselves are a form of data, and measures must be taken to prevent data leaks. Involving all stakeholders is crucial if we want to use AI for responsible decision-making, ensuring that AI models not only provide predictions but also explain why a certain choice is better and how confident the prediction is.