We use machine learning to study patients in translational settings. This includes integrative “multiomics” analysis across genomics, proteomics, imaging, and single-cell technologies, as well as quantitative clinical phenotyping using electronic health records and wearable devices. For more details on our work in immunology, mass cytometry, and clinical research see our sister laboratories (Gaudilliere Lab) and (Angst Lab).
Patient Phenotyping and Trajectory Prediction
The first step towards a precision medicine approach to healthcare is the development of holistic models that understand the current state of a given patient, and, can predict their future trajectory. Such models can then be used to guide future interventions. Our group combines data from Electronic Health Records, wearable devices, high-fidelity clinical measurements, and population-level databases to better understand long-term patient trajectories.
The immune system is involved in a symbiotic interplay with various biological and environmental factors. To understand the immune system in a clinical context these relationships must be characterized accurately. We develop integrative statistical models that bring together various datasets (including from the transcriptome, metablome, microbiombe, and proteome) of patients to both diagnose diseases more accurately and to understand their underlying biological mechanisms.
Flow/Mass Cytometry Bioinformatics
Flow/mass cytometry instruments analyze millions of cells per sample, creating a wide range of challenges for standard bioinformatics tools developed for genomics and proteomics. On the other hand, the large number of observations in these datasets (millions of measurements per patient) provides a unique environment for testing new machine learning concepts. The intersection of high-throughput single-cell biology, cytometry, and machine learning is a major focus of the lab.
Algorithm Benchmarking for High-throughput Single Cell Assays
Single cell biology is a young field. Shared datasets are scarce and empirical evaluation best practices are not yet adopted by the community. Old algorithms often out-perform new ones when evaluated independently. Users are not sure which algorithms should be used for which problem. FlowCAP is an international initiative to fix this problem. Together with DREAM we further extend this to the general systems biology community.
Model Reduction / Explainable Artificial Intelligence
Developing accurate predictive models is just a good start. Reproducing those same accurate results using a limited number of measurements is often essential for regulatory approval, to develop a viable product that would scale well, and to understand the biological processes drive the model’s predictive power. Stanford is home to the world’s leaders in sparse learning, post-selection inference, and other model reduction techniques (e.g., Rob Tibshirani and Trever Hastie) who help us tackle these challenges.
Human pregnancies last precisely nine months. To do that, mothers rely on a finely tuned immune balance (or immune clock), which controls an immunological switch from tolerance to rejection of the fetus (a hemi-allogeneic foreign body!). In collaboration with the March of Dimes and the Bill and Melinda Gates’ foundations, the lab studies the adaptation of feto-maternal immunity to normal and abnormal pregnancies (e.g. associated with preterm birth and preeclampsia). Using an multi-disciplinary approach that integrates deep immune profiling of pregnancy with changes in cell-free RNA expression (collaboration with Steve Quake), metabolomics (Mike Snyder) and the maternal microbiota (David Relman), our goal is to identify peripheral blood signatures that predict pregnancy complications.
Recovery from Trauma
How did the caveman survive after getting attacked by a saber toothed tiger? Turns out the immune system has evolved sophisticated cellular programs that enables us to heal and recover from major trauma. Using high dimensional mass cytometry, we have recently shown that the signaling behavior of a network of innate immune cells measured before surgery predicted surgical recovery. We are now in the process of prospectively connecting these predictions to measurements from wearable devices to provide patients with real-time feedback. Ongoing work in humans and animal models focuses pre-operative interventions (“pre-hab”) that alter a patient’s immune state to improve recovery after surgery.
Stroke is the number one cause of long-term disability in the world. Acute Stroke elicits a profound systemic inflammatory response, not unlike traumatic injury. We are interested in early immunological mechanisms, mobilized hours to days after the ischemic event that predict patients’ long-term neurocognitive recovery. This is a collaboration with Marion Buckwalter and Maarten G Lansberg.
Long-term space travel is around the corner! For thousands of years, our immune system has evolved here on earth, under 1g gravity. What happens when gravity disappears? Accumulating evidence suggests that astronauts immune system is depressed in space. This has both short and long term consequences to their health, with increased predisposition to opportunistic infection and perhaps cardiovascular disease. We are lucky to participate in multidisciplinary and collaborative efforts (including a project launched by Millie Hughes-Fullford, a former STS-40 Spacelab Life Sciences astronaut) to understand the adaptation of the human immune system to long-term exposure space travel.