Danish Researchers Launch PathogenFinder2 AI Tool to Predict Disease Potential of Unknown Bacterial Species

DTU researchers develop PathogenFinder2, an AI tool that identifies harmful bacteria before they infect humans, revolutionizing global pandemic preparedness.

By: AXL Media

Published: Mar 28, 2026, 5:19 AM EDT

Source: Information for this report was sourced from DTU (Technical University of Denmark)

A Proactive Shield Against Emerging Pathogens

The introduction of PathogenFinder2 marks a fundamental shift in epidemiological strategy, moving from reactive containment to proactive identification. Developed by a multi-institutional team at the DTU National Food Institute, the AI tool is designed to scan the genetic architecture of bacteria found in sewage, wildlife, and healthy humans to flag those with the potential to cause disease. By identifying these "unseen" threats before the first documented human infection, health authorities can gain a critical head start in developing diagnostic tests and vaccines, potentially preventing localized outbreaks from escalating into global health crises.

Leveraging Protein Language Models for Genomic Interpretation

Unlike traditional computational methods that rely on comparing new bacteria to a database of known relatives, PathogenFinder2 utilizes a sophisticated protein language model. This advanced AI is trained on millions of protein sequences, allowing it to interpret the "language" of microbial biochemistry rather than simple genetic similarity. This strategy is particularly effective for novel species that lack close documented relatives, enabling the system to detect subtle biochemical signals of virulence that conventional software would overlook. Researcher Alfred Ferrer Florensa noted that this model is among the first to interpret whole bacterial genomes by leveraging the massive predictive potential of these language systems.

Navigating the Complexities of Microbial Diversity

The challenge of identifying risky bacteria has intensified as climate change and ecosystem expansion bring researchers into contact with previously undocumented species. While the vast majority of bacteria are either harmless or beneficial to human health, the small fraction capable of causing serious infection is becoming harder to isolate through traditional laboratory experiments, which are often slow and prohibitively expensive. PathogenFinder2 addresses this bottleneck by providing a rapid, digital assessment of bacterial pathogenic capacity, allowing scientists to prioritize which organisms require urgent physical validation in a controlled lab setting.