A high-content screening combined with AI-assisted vibe coding approach identifies a mitochondria-to-lysosome signaling cascade ameliorating Parkinson’s pathology

The emergence of vibe coding, a natural language-driven programming paradigm powered by large language models, is transforming computational biology by enabling researchers without extensive coding expertise to design and execute complex analyses. Leveraging high-throughput imaging integrated with vibe coding workflows, we screened a natural compound library and identified the triterpenoid betulinaldehyde (Betu) as a potent mTOR-independent activator of TFEB. In cross-species Parkinson’s disease (PD) models, including patient-derived dopaminergic neurons, transgenic Drosophila and mice, Betu rescued neurodegeneration and restored lysosomal function and autophagic flux. Mechanistically, Betu induced mild mitochondrial permeability transition pore (mPTP) opening, triggering Ca 2+ efflux that activated calcineurin and promoted TFEB nuclear translocation, thereby enhancing lysosomal biogenesis and function. The anti-aging lipid ceramide, a known inducer of moderate mPTP opening, similarly engaged this mitochondria–lysosome axis, suggesting a general cytoprotective response to cellular stress and aging through mitochondria-to-lysosome communication. Collectively, these findings demonstrate how large language model-assisted vibe coding democratizes computational biology by enabling resource-limited, non-programming scientists to perform high-throughput, reproducible analyses, while identifying a novel mitochondria-to-lysosome signaling cascade (mitochondria–Ca 2+ /calcineurin/TFEB/lysosome biogenesis) as a therapeutic target for PD.