A collaborative team from Sichuan University and Huazhong University of Science and Technology has jointly developed an AI system dubbed ‘Xunzi.’ Built on a substantial model boasting 7.3 billion parameters and honed through two rounds of focused training, the system attains an impressive 86% accuracy rate in evaluating biomedical gene-disease associations, outperforming GPT-4o’s 65%.
‘Xunzi’ employs logical reasoning and multi-omics data integration to screen disease targets. In Parkinson’s disease research, it has pinpointed kinases like CHK2 and, through cellular and mouse experiments, confirmed that inhibiting CHK2 enhances motor function in mice and boosts the survival rate of dopamine neurons. The study also uncovered a novel functional link between CHK2 and LRRK2, a pivotal genetic factor in Parkinson’s disease.
Moreover, the team spearheaded by Cao Nan from Sun Yat-sen University harnessed ‘Xunzi’ for heart regeneration research, substantially reducing target screening time and amplifying research efficiency. Nevertheless, the current dependability of ‘Xunzi’ in predicting rare diseases requires enhancement, and challenges such as AI hallucinations and biases in manual verification may persist. Looking ahead, the team intends to incorporate more data and link up with automated experimental platforms to further refine the system.