待翻译：China releases powerful DNA-screening AI tool free to help fight rare diseases

AI 服务暂时不可用，以下为来源正文，待恢复后补全翻译。

For people navigating rare diseases, which are mostly genetic in origin, the search for answers is a gruelling marathon often called the “diagnostic odyssey”. Patients can spend years visiting specialists and undergoing inconclusive tests, while the underlying cause of their condition remains hidden within three billion base pairs of DNA in the human genome. But Artificial intelligence is emerging as a tool to shorten this time frame. Chinese researchers have created an open-source AI system capable of bridging the gap between raw DNA sequences and the world’s vast library of clinical literature to formulate treatment approaches. According to one of its developers, BGI-Research, OneGenome has outperformed general large language models such as DeepSeek-v4 and other traditional gene models in multiple clinical diagnostic and medication guidance tests. Rather than reading DNA sequences, OneGenome is trained to reason. It uses accumulated human medical logic to interpret the clinical consequences of gene mutations. It was developed by integrating Genos – a genomic foundational model released last year that was trained on human genomes to represent diverse global populations – with large language model capabilities. “Unlike common diseases, rare conditions suffer from a lack of knowledge, making an accurate diagnosis very challenging,” Liu Shiping, chief scientist in brain science at BGI-Research, said in an interview. “This is our attempt to accelerate that timeline. After building a comprehensive database constructed from rare diseases, patients’ gene sequencing data and clinical manifestations, we can deploy AI to analyse these findings,” he said of OpenRare, an open-source genetic analysis system for rare disease diagnosis. “In the future, AI can support doctors in identifying potential disease-causing mutations so that they no longer have to rely only on their experience to investigate each possibility one by one.” The automation of screening and clinical reasoning enables the model to compress a process that used to take years into just a few hours, while also generating a traceable analysis. The goal is to provide doctors with leads that will be further verified and eventually help establish a diagnosis, Liu said. The team also applied the AI model to treat rare types of cancer by identifying therapeutic targets for developing personalised immunotherapy. At the United Nations’ AI for Good Global Summit 2026 in July, where the team received the ITU Innovate for Impact Use Case Award, the team demonstrated the model with one case of a desmoplastic small round cell tumour, a type of rare and aggressive soft-tissue cancer. The OneGenome process pinpointed 11 therapeutic targets that standard clinical testing had overlooked in a late-stage patient. Liu said the traditional approach of treating each mutation as an isolated actor failed to capture how these mutations interacted with each other. “Genomic sequences are connected instead of independent. When a sequence is removed, the whole network could change,” Liu explained. Liu’s vision for the technology was to turn it into a global public good, especially for developing countries where medical resources were limited. He said he hoped that in the future, “AI tools will be able to help individuals analyse their health data so patients can communicate more effectively with doctors to confirm diagnosis”. Passionate about science? Dive deeper with the Dark Matters newsletter, a weekly in-depth analysis on China’s rise in science, technology and military that goes under the surface. Sign up for free now.