Monthly Tutorial: Sebastian Cruz

Title: A Learned Representation of the Mesenchymal Stromal Cell-Extracellular Vesicle Surface to Inform Engineered Nanoparticles

Abstract: Mesenchymal stromal cell extracellular vesicles (MSC-EVs) are among the leading cell-free candidates for anti-inflammatory therapy, with immune cells serving as their primary recipients in vivo. Among these, macrophages are key target cells, yet the surface ligands that mediate EV docking and uptake remain poorly characterized. Identifying these ligands is a critical step toward engineering nanoparticles with enhanced cell specificity and therapeutic potency. Progress has been limited by the sparsity and annotation bias of public EV proteomics datasets, which create a positive-unlabeled (PU) learning setting in which a small number of validated ligand–receptor interactions are vastly outnumbered by unlabeled surface proteins. To address this limitation, we present a computational pipeline that harmonizes community-sourced EV and proteomics data from 35 independent human MSC-EV studies aggregated across ExoCarta, Vesiclepedia, PRIDE, and MassIVE, together with macrophage receptor context from the Human Protein Atlas. From this integration we rank 313 MSC-EV surface ligands and 1,274 curated ligand–receptor pairs by their strength of interaction with macrophage receptors through a composite engagement index and a bagged PU classifier. Our preliminary results show that the MSC-EV surface interactome is dominated by opsonins, efferocytosis ligands, and matricellular ECM proteins engaging β1 integrins and LRP receptors rather than by classical soluble cytokines. PU-prioritized candidates occupy a similar engagement geometry as held-out anchors in UMAP and canonical correlation analysis (CCA) space, and the classifier recovers 23 of 41 strictly bone-marrow MSC-EV anchor genes within the top-ranked band, providing evidence of alignment with a coherent molecular program relevant to immune-cell uptake of EVs. These findings suggest a reframing of MSC-EV/macrophage interaction as an adhesion and phagocytic recognition code and point to a route by which ML-informed surface biology can guide next-generation nanoparticle design to mimic natural EV delivery.

Speaker Bio: Sebastián A. Cruz Romero is a Computer Science & Engineering graduate and current Masters student in Bioengineering at the University of Puerto Rico at Mayagüez. He combines computational modeling with functional validation in Dr. Maribella Domenech’s Micro-Biomechanics Research Laboratory to characterize the surface and molecular program that drives MSC-derived extracellular vesicles to engage macrophages for enhanced immune-cell targeting and potency. He is also co-founder and CEO of Capicú Technologies, a Puerto Rico–based applied AI/ML startup that deploys compressed models directly on instruments to monitor critical process parameters and quality attributes, and automate downstream analysis in-place, without the compute infrastructure that frontier AI models demand. His research sits at the intersection of representation learning, computational biology, and model compression. Sebastián has committed to building a deep-tech culture rooted in Puerto Rico, to deal with health disparities and biological questions that remain poorly addressed in his community.