Chinese scientists have developed a specialized artificial intelligence system capable of searching for disease-causing mutations and identifying their mechanisms of influence in more than 5,800 genetic diseases.
This system helped scientists discover disruptions in the function of two genes linked to the development of Parkinson's disease, according to a scientific article published in the journal Nature Biomedical Engineering.
The study stated: "Hypothesis generation and verification in biomedicine remains limited, as humans struggle to integrate fragmented data from different scientific fields and discover new patterns within them. We have developed an 'artificial biologist' capable of making logical judgments and generating fully and automatically testable hypotheses when analyzing a large number of experimental results."
This system was developed by a team of Chinese mathematicians and biologists led by Professor Jia Da from Sichuan University. It combines two key features of large-scale linguistic models and specialized artificial intelligence systems for scientific information processing: the ability to construct logical chains and interpret data, along with the skills of integrating heterogeneous data characteristic of specialized machine learning systems.
To train the "artificial biologist," the researchers compiled a massive amount of scientific publications and data, including more than 24.4 million peer-reviewed scientific articles and over 613 terabytes of scientific data. This data included information on the structure and interactions of more than 21,000 human protein-coding genes, along with their potential associations with approximately 5,850 diseases.
The scientists explained that they divided the AI system into two parts, similar to how the human brain works: one part constructs logical chains, while the other handles holistic thinking and big-picture analysis. According to the researchers, this design helped reduce what are known as "AI hallucinations" and inaccurate responses, and made the system more capable of analyzing data compared to several advanced large-scale language models.
Initial tests showed that the system is capable of searching for genes associated with non-small cell lung cancer with greater accuracy than some other neural algorithms. It also helped scientists discover a possible link between the development of Parkinson's disease and disruptions in the function of two genes, CHK2 and IRAK4, as restoring their activity in mouse studies led to a reduction in some of the disease's symptoms.
The researchers concluded that these results point to promising prospects for the use of "artificial biology" in scientific and medical research in the future.
