Where immunology, machine learning and medicine meet.
Four connected themes, from T cells inside tumors to models that read single-cell and clinical data. Each lists the methods involved and selected works I have contributed to.
Fig. 02.1Cells in a learned embedding, linked by a nearest-neighbor graph
01/ 049 works
Machine learning for single-cell & spatial omics
Implementing and benchmarking deep-learning models — graph neural networks and transformers — for single-cell and spatial omics, with contributions to an integrated cell atlas of endoderm-derived organoids, a benchmark of feature selection for single-cell data integration and a study of RNA outside segmented cells in spatial transcriptomics.
Fig. 02.2A cytotoxic T cell’s track through tumor tissue to its target
02/ 048 works
Tumor immunology & cell therapy
How cytotoxic T cells navigate and survive inside tumors — from CAR T-cell therapy of solid tumors in preclinical models and intravital imaging of T-cell–myeloid interactions to single-cell and spatial maps of the tumor microenvironment.
Fig. 02.3A peptide in the HLA groove, with one polymorphic residue
03/ 041 work
Transplant immunology & hematologic malignancies
Extending a pipeline that predicts minor histocompatibility antigens from exome and transcriptome data of allogeneic stem-cell transplant cohorts, with mitochondrial and nuclear variant calling for chimerism and mosaicism analysis.
Fig. 02.4A clinical record read token by token, attending to key entities
04/ 043 works
AI for clinical decision-making
Testing where large language models can support oncology care — from treatment recommendations for tumor-board cases to plain-language surgical reports for patients — and learning shared representations of health across diseases, modalities and time.