Add {obs,var}_names_make_unique! (#30)

* add counterpart of Python AnnData's var_names_make_unique

* test counterpart of Python AnnData's var_names_make_unique

* document counterpart of Python AnnData's var_names_make_unique

* add/test/doc counterpart of Python AnnData's obs_names_make_unique

* try to fix Logging compat

* try to fix make_unique docs examples

* fix: typo in duplicateindices

* fixup! fix: typo in duplicateindices

---------

Co-authored-by: jdm204 <[email protected]>

Project.toml

@@ -23,13 +23,15 @@ CompressHashDisplace = "0.1.2"
 DataFrames = "0.22, 1"
 FileIO = "1.6"
 HDF5 = "0.16 - 0.99, 1"
+Logging = "1.6.7"
 OrderedCollections = "1.6"
 PooledArrays = "1"
 StructArrays = "0.6.4"
 julia = "1.5"
 
 [extras]
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
+Logging = "56ddb016-857b-54e1-b83d-db4d58db5568"
 
 [targets]
-test = ["Test"]
+test = ["Test", "Logging"]

docs/src/objects.md

@@ -60,6 +60,22 @@ ad_sub[[true,false,true],:]
 ad_sub[[1,3],:]
 ```
 
+### Operations on AnnData Objects
+
+Make duplicate `var_names` unique by appending a numbered suffix.
+
+```@example 1
+import Muon: var_names_make_unique! # hide
+var_names_make_unique!(ad)
+```
+
+Similarly, we can make `obs_names` unique also.
+
+```@example 1
+import Muon: obs_names_make_unique! # hide
+obs_names_make_unique!(ad)
+```
+
 ## 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/Muon.jl

@@ -15,6 +15,7 @@ using FileIO
 
 export readh5mu, readh5ad, writeh5mu, writeh5ad, isbacked, update_obs!, update_var!, update!
 export AnnData, MuData
+export var_names_make_unique!, obs_names_make_unique!
 
 import Pkg
 # this executes only during precompilation

src/util.jl

@@ -114,3 +114,78 @@ Base.firstindex(A::Union{AbstractMuData, AbstractAnnData}, d::Integer) = 1
 Base.lastindex(A::Union{AbstractMuData, AbstractAnnData}, d::Integer) = size(A, d)
 
 Base.copy(d::Union{MuDataView, AnnDataView}) = parent(d)[parentindices(d)...]
+
+"""
+    var_names_make_unique!(A::AnnData, join = '-')
+
+Make `A.var_names` unique by appending `join` and sequential numbers
+(1, 2, 3 etc) to duplicate elements, leaving the first unchanged.
+"""
+function var_names_make_unique!(A::AnnData, join='-')
+    index_make_unique!(A.var_names, join)
+end
+
+"""
+    obs_names_make_unique!(A::AnnData, join = '-')
+
+Make `A.obs_names` unique by appending `join` and sequential numbers
+(1, 2, 3 etc) to duplicate elements, leaving the first unchanged.
+"""
+function obs_names_make_unique!(A::AnnData, join='-')
+    index_make_unique!(A.obs_names, join)
+end
+
+function index_make_unique!(index, join)
+    duplicates = duplicateindices(index)
+
+    if isempty(duplicates)
+        @info "var names are already unique, doing nothing"
+        return nothing
+    end
+
+    example_colliding_names = []
+    set = Set(index)
+
+    for (name, positions) in duplicates
+        i = 1
+        for pos in Iterators.rest(positions, 2)
+            while true
+                potential = string(index[pos], join, i)
+                i += 1
+                if potential in set
+                    if length(example_colliding_names) <= 5
+                        push!(example_colliding_names, potential)
+                    end
+                else
+                    index[pos] = potential
+                    push!(set, potential)
+                    break
+                end
+            end
+        end
+    end
+
+    if !isempty(example_colliding_names)
+        @warn """
+              Appending $(join)[1-9...] to duplicates caused collision with another name.
+              Example(s): $example_colliding_names
+              This may make the names hard to interperet.
+              Consider setting a different delimiter with `join={delimiter}`
+              """
+    end
+end
+
+function duplicateindices(v::Muon.Index{T, I}) where {T <: AbstractString, I <: Integer}
+    varnames = Dict{T, Vector{Int64}}()
+
+    for i in eachindex(v)
+        if haskey(varnames, v[i])
+            push!(varnames[v[i]], i)
+        else
+            varnames[v[i]] = [i]
+        end
+    end
+
+    filter!(x -> length(last(x)) > 1, varnames)
+    varnames
+end

test/anndata.jl

@@ -39,3 +39,25 @@ end
     @test copy(adview).X == subad.X
     test_ad_slicing(subad, 50, 5, x[i, 3:7])
 end
+
+@testset "unique names" begin
+    @test_logs (:info,) var_names_make_unique!(ad)
+    @test_logs (:info,) obs_names_make_unique!(ad)
+    ad2 = deepcopy(ad)
+    ad2.var_names[3] == "10"
+    ad2.obs_names[90] == "obs_30"
+    var_names_make_unique!(ad2)
+    obs_names_make_unique!(ad2)
+    @test allunique(ad2.var_names)
+    @test allunique(ad2.obs_names)
+    ad2.var_names[10] = "10-1"
+    ad2.var_names[3] = "10"
+    ad2.var_names[4] = "10"
+    ad2.obs_names[11] = "obs_10-1"
+    ad2.obs_names[10] = "obs_10"
+    ad2.obs_names[9] = "obs_10"
+    @test_logs (:warn,) var_names_make_unique!(ad2)
+    @test_logs (:warn,) obs_names_make_unique!(ad2)
+    @test allunique(ad2.var_names)
+    @test allunique(ad2.obs_names)
+end

test/runtests.jl

@@ -1,4 +1,4 @@
-using Muon, Test
+using Muon, Test, Logging
 
 @testset "Index" begin
     include("index.jl")