Add DataFrame constructor for AnnData

Author: jdm204

docs/Project.toml

@@ -1,3 +1,4 @@
 [deps]
+DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
 Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"
 Documenter = "e30172f5-a6a5-5a46-863b-614d45cd2de4"

docs/make.jl

@@ -1,4 +1,4 @@
-using Documenter, Muon
+using Documenter, Muon, DataFrames
 
 makedocs(sitename="Muon Documentation", warnonly=:cross_references)
 

docs/src/objects.md

@@ -76,6 +76,21 @@ import Muon: obs_names_make_unique! # hide
 obs_names_make_unique!(ad)
 ```
 
+The data matrices of `AnnData` objects can be converted to a `DataFrame`, annotated with `obs` and `var` names.
+
+```@example 1
+using DataFrames
+DataFrame(ad)
+```
+
+By default, the first column `obs` corresponds to the `obs_names` and the remaining columns are named according to the `var_names`. To obtain the transpose of this, pass `columns=:obs`.
+
+To use a different data matrix (the default is `ad.X`), pass the name of the layer:
+
+```julia
+DataFrame(ad, layer="raw")
+```
+
 ## MuData
 
 The basic idea behind a multimodal object is _key_ ``\rightarrow`` _value_ relationship where _keys_ represent the unique names of individual modalities and _values_ are `AnnData` objects that contain the correposnding data. Similarly to `AnnData` objects, `MuData` objects can also contain rich multimodal annotations.

src/util.jl

@@ -189,3 +189,35 @@ function duplicateindices(v::Muon.Index{T, I}) where {T <: AbstractString, I <:
     filter!(x -> length(last(x)) > 1, varnames)
     varnames
 end
+
+"""
+    DataFrame(A::AnnData; layer=nothing, columns=:var)
+
+Return a DataFrame containing the data matrix `A.X` (or `layer` by
+passing `layer="layername"`). By default, the first column contains
+`A.obs_names` and the remaining columns are named according to
+`A.var_names`, to obtain the transpose, pass `columns=:obs`.
+"""
+function DataFrames.DataFrame(A::AnnData; layer::Union{String, Nothing}=nothing, columns=:var)
+    if columns ∉ [:obs, :var]
+        throw(ArgumentError("columns must be :obs or :var (got: $columns)"))
+    end
+    rows = columns == :var ? :obs : :var
+    colnames = getproperty(A, Symbol(columns, :_names))
+    if !allunique(colnames)
+        throw(ArgumentError("duplicate column names ($(columns)_names); run $(columns)_names_make_unique!"))
+    end
+    rownames = getproperty(A, Symbol(rows, :_names))
+
+    M = if isnothing(layer)
+        A.X
+    elseif layer in keys(A.layers)
+        A.layers[layer]
+    else
+        throw(ArgumentError("no layer $layer in adata layers"))
+    end
+    df = DataFrame(columns == :var ? M : transpose(M), colnames)
+    setproperty!(df, rows, rownames)
+    select!(df, rows, All())
+    df
+end

test/anndata.jl

@@ -61,3 +61,17 @@ end
     @test allunique(ad2.var_names)
     @test allunique(ad2.obs_names)
 end
+
+@testset "DataFrame conversion" begin
+    using DataFrames
+    df = DataFrame(ad)
+    @test names(df) == ["obs"; ad.var_names]
+    @test df.obs == ad.obs_names
+    ad.var_names[3] = "10"
+    @test_throws ArgumentError DataFrame(ad)
+    @test_throws ArgumentError DataFrame(ad, columns=:foo)
+    @test_throws ArgumentError DataFrame(ad, layer="doesn't exist")
+    df2 = DataFrame(ad, columns=:obs)
+    @test names(df2) == ["var"; ad.obs_names]
+    @test df2.var == ad.var_names
+end