AI and the World's Hard Problems · Life sciences · Nature (computational biology) · 2026-10-21
Spatial metabolomics-guided deep learning framework for dissecting immune evasion mechanisms in fungal infection microenvironments
The work targets the spatially organized metabolic circuits that let fungal pathogens evade host immunity in invasive infections, which cause high mortality in immunocompromised patients. Coupling high-resolution spatial metabolomics of murine invasive pulmonary aspergillosis tissue with an attention-enhanced multi-scale convolutional network, the framework segments the infection core, immune infiltration zone, and transitional boundary zone with a mean Dice coefficient of 0.887 against expert pathologist annotations. A variational graph autoencoder with neural ordinary differential equations then infers how metabolite–immune interaction networks evolve over 24–72 hours, ranking kynurenine as the highest-centrality hub metabolite and placing a putative metabolic tipping point between 36 and 48 hours.