Build reference for consensus cell type labels

analyses/cell-type-consensus/README.md

@@ -5,11 +5,30 @@ Specifically, the cell type annotations obtained from both `SingleR` and `CellAs
 
 ## Description
 
-TBD
+The goal of this module is to create a reference that can be used to define an ontology aware consensus cell type label for all cells across all ScPCA samples. 
+This module performs a series of steps to accomplish that goal: 
+
+1. The cell type annotations present in the `PanglaoDB` reference file were assigned to an ontology term identifier, when possible.
+See [`references/README.md`](./references/README.md) for a full description on how we completed assignments.  
+2. We looked at all possible combinations of cell type labels between the `PanglaoDB` reference (used with `CellAssign`) and the `BlueprintEncodeData` reference (used with `SingleR`). 
+We then explored using a set of rules used to define consensus cell types in [`exploratory-notebooks/01-reference-exploration.Rmd`](./exploratory-notebooks/01-reference-exploration.Rmd). 
+3. We created a [reference table](./references/consensus-cell-type-reference.tsv) containing all combinations for which we were able to identify a consensus cell type label.  
+The consensus cell type label corresponds to the [latest common ancestor (LCA)](https://rdrr.io/bioc/ontoProc/man/findCommonAncestors.html) between the `PanglaoDB` and `BlueprintEncodeData` terms. 
+
+When creating the consensus cell type labels we implemented the following rules: 
+
+- If the terms share more than 1 LCA, no consensus label is set. 
+The only exception is if one of the LCA terms corresponds to `hematopoietic precursor cells`. 
+If that is the case all other LCA terms are removed and `hematopoietic precursor cell` is used as the consensus label. 
+- If the LCA has greater than 170 descendants, no consensus label is set, with some exceptions: 
+  - When the LCA is `neuron`, `neuron` is used as the consensus label. 
+  - When the LCA is `epithelial cell` and the annotation from `BlueprintEncodeData` is `Epithelial cells`, then `epithelial cell` is used as the consensus label. 
+  - If the LCA is `bone cell`, `lining cell`, `blood cell`, `progenitor cell`, or `supporting cell`, no consensus label is defined. 
+
 
 ## Usage
 
-TBD
+See the [`scripts/README.md`](./scripts/README.md) for instructions on running the scripts in this module. 
 
 ## Input files
 

analyses/cell-type-consensus/references/README.md

@@ -30,3 +30,19 @@ There were no terms that encompassed both other than `progenitor cell`.
 Monocytes differentiate into mononuclear osteoclasts which are then activated and become multinucleated osteoclasts.
 Because monocytes are the "precursor" to the differentiated osteoclast, we chose to use this term.
 - `NA` was used for `Undefined placental cells` and `Transient cells` as no clear cell type from the cell ontology was identified.
+
+2. `consensus-cell-type-reference.tsv`: This file contains a table with all cell type combinations between the `PanglaoDB` reference and `BlueprintEncodeData` reference for which a consensus cell type is identified. 
+
+The table includes the following columns: 
+
+|  |   |
+| --- | --- |
+| `panglao_ontology` | Cell type ontology term for `PanglaoDB` cell type |
+| `panglao_annotation` | Original name for the cell type as set by `PanglaoDB` |
+| `blueprint_ontology` | Cell type ontology term for `BlueprintEncodeData` cell type |
+| `blueprint_annotation_main` | Original name for the cell type as set by `BlueprintEncodeData` (main term) |
+| `blueprint_annotation_fine` | Original name for the cell type as set by `BlueprintEncodeData` (fine term) |
+| `consensus_ontology` | Cell type ontology term for consensus cell type |
+| `consensus_annotation` | Human readable name for the consensus cell type |
+
+This file was generated by running [`scripts/02-prepare-consensus-reference.R`](../scripts/02-prepare-consensus-reference.R).