# CellaRepertorium <!-- badges: start --> [](https://github.com/amcdavid/CellaRepertorium/actions) <!-- badges: end --> This package contains methods for clustering, pairing and testing single cell RepSeq data, especially as generated by [10X genomics VDJ solution](https://support.10xgenomics.com/single-cell-vdj). ## Installation Install with install.packages('BiocManager') # if you don't have it yet BiocManager::version() # Check that Bioconductor version >= 3.12 BiocManager::valid() # And it's in a sane state. BiocManager::install('CellaRepertorium') # install For the development version, install via `devtools::install_github('amcdavid/CellaRepertorium')`. ## Data requirements and package structure The fundamental unit this package operates on is the **contig**, which is a section of contiguously stitched reads from a **single cell**. Each contig belongs to one (and only one) cell, however, cells generate multiple contigs. <img src = man/figures/contig_schematic.png /> Contigs can also belong to a **cluster**. Because of these two many-to-one mappings, these data can be thought as a series of ragged arrays. The links between them mean they are relational data. A `ContigCellDB()` object wraps each of these objects as a sequence of three `data.frames` (`dplyr::tibble()`, actually). `ContigCellDB()` also tracks columns (the primary keys) that uniquely identify each row in each of these tables. The `contig_tbl` is the `tibble` containing **contigs**, the `cell_tbl` contains the **cells**, and the `cluster_tbl` contains the **clusters**. The `contig_pk`, `cell_pk` and `cluster_pk` specify the columns that identify a contig, cell and cluster, respectively. These will serve as foreign keys that link the three tables together. The tables are kept in sync so that subsetting the contigs will subset the cells, and clusters, and vice-versa. <img src = man/figures/table_schematic.png /> Of course, each of these tables can contain many other columns that will serve as covariates for various analyses, such as the **CDR3** sequence, or the identity of the **V**, **D** and **J** regions. Various derived quantities that describe cells and clusters can also be calculated, and added to these tables, such as the **medoid** of a cluster – a contig that minimizes the average distance to all other clusters. ## Some functions of interest Mainly, this package seeks to enforce proper schema of single cell repertoire data and stay out the user’s way for various summaries they might conduct. However, there are a variety of specialized functions, as well: - `cdhit_ccdb()`: An R interface to CDhit, which was originally ported by Thomas Lin Pedersen. - `fine_clustering()`: clustering CDR3 by edit distances (possibly using empirical amino acid substitution matrices) - `canonicalize_cell()`: Return a single contig for each cell, e.g., for combining VDJ information with 5’-based single cell expression - `ccdb_join()`: join a `ccdb` object from this package to a `SingleCellExperiment` object, by droplet barcode. - `cluster_permute_test()`: permutation tests of cluster statistics - `cluster_logistic_test()`: logistic regression tests for overrepresentation of clusters among cells - `pairing_tables()`: Generate pairings of contigs within each cell in a way that they can be plotted ## Interfacing related packages for clonal analyses - To combine repertoire information with expression of endogenous mRNAs, this package has been used with `SingleCellExperiment::SingleCellExperiment()` and [Seurat](https://satijalab.org/seurat/) after generating cell canonicalizations. - Many tools from the [Immcantation](https://alakazam.readthedocs.io/en/version-0.2.11/) suite can work directly on `ContigCellDB()` objects. # Acknowledgments Development of CellaRepertorium was funded in part by UL1 TR002001 (PI Bennet/Zand) pilot to Andrew McDavid.