Symphony

Symphony maps single-cell query datasets onto large-scale integrated reference atlases to enable rapid identification of cell states, reproducible annotation transfer, and integration of multimodal, multi-donor, and multi-species single-cell data.

Key Features:

• Large-scale integrated reference atlases: Supports mapping to multimillion-cell integrated reference atlases.
• Efficient query mapping: Maps query cells onto reference atlases within seconds.
• Stable low-dimensional reference embedding: Uses a stable low-dimensional embedding that localizes query cells while preserving spatial relationships and intrinsic structure.
• Reproducible annotation transfer: Enables reproducible transfer of annotations from the reference atlas to query datasets.
• Multimodal and cross-platform integration: Handles multimodal data such as CITE-seq and diverse sequencing platforms across multiple donors, species, disease states, and developmental stages.
• Protein expression inference: Enables inference of surface protein expression via mapping to multimodal CITE-seq atlases (e.g., memory T cell atlas).

Scientific Applications:

• Predicting cell types: Maps multi-donor, multi-species query data to identify pancreatic cell types accurately.
• Developmental trajectory analysis: Localizes cells along developmental trajectories, exemplified by tracing fetal liver hematopoiesis.
• Inference of surface protein expression: Infers surface protein expression using a multimodal CITE-seq atlas of memory T cells.

Methodology:

Constructs large-scale, portable reference atlases; employs a stable low-dimensional reference embedding to localize query cells; and uses embedding and mapping algorithms to align queries to references and enable annotation transfer.