The August Affinity Group hosts Zinaida Good, PhD — Assistant Professor of Medicine in the Divisions of Immunology and Rheumatology and of Computational Medicine at Stanford University, and Director of the Stanford Center for Cancer Cell Therapy Data Hub. Her talk, “Artificial intelligence systems to advance engineered T cell immunotherapy designs,” covers the computational program her lab has built around cell therapy — and features SAGE-FM, a spatial transcriptomics foundation model on which she is co-senior author.

Good’s research program aims to understand and enhance engineered T cell immunotherapies for cancer and immune-mediated diseases through computational approaches and systems immunology. Her lab combines machine learning with clinical multiomic datasets to build AI systems for advanced T cell therapy design — work that depends on being able to read cellular organization within a tumor, not just its average expression profile.

That is the gap SAGE-FM addresses. Spatial transcriptomics profiles gene expression while preserving tissue architecture, which motivates models that can capture spatially conditioned regulatory relationships rather than treating each spot in isolation. SAGE-FM is a lightweight foundation model built on graph convolutional networks and trained with a masked-central-spot prediction objective — the spatial analogue of masked-token pretraining. Trained on 416 human Visium samples spanning 15 organs, it learns spatially coherent embeddings that recover masked genes robustly, with 91% of masked genes showing significant correlations (p < 0.05).

The results are notable for what they suggest about model size. SAGE-FM’s embeddings outperform MOFA in unsupervised clustering and in preserving biological heterogeneity, and the model generalizes to downstream tasks — 81% accuracy on pathologist-defined spot annotation in oropharyngeal squamous cell carcinoma, and improved glioblastoma subtype prediction relative to MOFA. In silico perturbation experiments show the model captures directional ligand–receptor and upstream–downstream regulatory effects consistent with ground truth. Together these argue that simple, parameter-efficient GCNs can serve as biologically interpretable and spatially aware foundation models for large-scale spatial transcriptomics — a useful counterweight to the assumption that a foundation model has to be large to be good.

Good completed her PhD in Computational & Systems Immunology with Garry Nolan and Sean Bendall, and postdoctoral training on CAR T cell therapies with Crystal Mackall and Sylvia Plevritis, all at Stanford. Her research is supported by the NIH/NCI Pathway to Independence Award, the NIH/OD Multimodal AI Initiative Award, an NIH/NCI Program Project Grant, and the Weill Cancer Hub West. She was named an Arthur & Sandra Irving Cancer Immunology Fellow in 2022, a Parker Bridge Fellow in 2023, and an AACR–Women in Cancer Research Scholar in 2024.

The talk is Friday, August 7, 2026 at 9:00 AM Hawaiʻi time, on Zoom.

Register for this Affinity Group meeting →

Register once and Zoom will email you the join link plus reminders for this and any future Affinity Group meeting. Recordings are posted to the AI PHI YouTube channel ↗. To suggest a future speaker, email aiphi@shepherdresearchlab.org.