PhEMD

"PhEMD" (Phenotypic Earth Mover's Distance) is a computational method to identify axes of variation among multicellular biospecimens profiled at single-cell resolution. Recognizing the gap in tools for analyzing variation among complex biospecimens rather than individual cells, PhEMD introduces a unique approach by embedding a "manifold of manifolds," where each datapoint in the higher-level manifold represents a collection of points within a lower-level manifold of cells.

Key Features and Functionalities:

- Identification of Axes of Variation: PhEMD is tailored to uncover axes of subpopulational variation among biospecimens under various perturbation conditions, enriching our understanding of complex biological and clinical phenomena.

- Inference of Unprofiled Phenotypes: A distinctive feature of PhEMD is its ability to infer the phenotypes of biospecimens that have not been directly profiled, providing valuable insights into potential biological states based on existing data.

- Application to Clinical Datasets: When applied to clinical datasets, PhEMD can generate a map of patient-state space, illuminating sources of variation from one patient to another and potentially uncovering novel aspects of disease mechanisms and patient responses.

- Scalability and Compatibility: PhEMD is designed to be scalable to large datasets, ensuring its applicability in high-throughput experiments. It is also compatible with leading batch-effect correction techniques, making it versatile across different experimental designs.

- Generalizability: The method is generalizable to multiple experimental designs, offering flexibility for researchers in diverse biological and clinical research fields.