implement generic inchitomol() with stereo support

[hhaensel]

This PR solves #137 and supports parsing of stereo information.

julia> mol1 = smilestomol("[D]O[D]")
{3, 2} simple molecular graph SMILESMolGraph

julia> i1 = inchi(mol1)
"InChI=1S/H2O/h1H2/i/hD2"

julia> inchitomol(i1)
{3, 2} simple molecular graph SDFMolGraph

julia> mol2 = smilestomol("C[C@H](O)C(=O)O")
{7, 6} simple molecular graph SMILESMolGraph

julia> i2 = inchi(mol2)
"InChI=1S/C3H6O3/c1-2(4)3(5)6/h2,4H,1H3,(H,5,6)/t2-/m0/s1"

julia> inchitomol(i2)
{7, 6} simple molecular graph SDFMolGraph

julia> inchitomol(i2, stereo = false)
{7, 6} simple molecular graph SDFMolGraph

[hhaensel]

Something still needs a fix. The demomol from the inchi test fails

[mojaie]

I haven't checked it very well yet, but I think it's wonderful. Thank you very much.
I'm considering resolving the isotope issue by creating simple aliases like the following. Is that okay?

SDFAtom(; sym=:D, kwargs...) = SDFAtom(; sym=:H, isotope=2, kwargs...)

[hhaensel]

If we'd go that way, we would need to adapt the drawing routines appropriately, see e.g.

drawsvg(smilestomol("[D]O[2H]"))

We might introduce a switch to distiinguish whether we want '2H' or 'D' to be displayed.
For my application I have left everything as is, and implemented routines to exchange D with 2H

function replace_atom(x::T, symbol::Symbol) where T
    values = [field == :symbol ? symbol : getfield(x, field) for field in fieldnames(T)]
    T(values...)
end

function replace_atom(x::T, symbol::Symbol, isotope::Union{Nothing,Int}) where T
    values = [field == :symbol ? symbol : field == :isotope ? isotope : getfield(x, field) for field in fieldnames(T)]
    T(values...)
end

# write all deuterium atoms as '2H'
function d_to_h!(mol)
    for (k, v) in mol.vprops
        if v.symbol == :D
            mol.vprops[k] = replace_atom(v, :H, 2)
        end
    end
    mol
end

# write all deuterium atoms as 'D'
function h_to_d!(mol)
    for (k, v) in mol.vprops
        if v.symbol == :H && v.isotope == 2 || v.symbol == :D && v.isotope != 2
            mol.vprops[k] = replace_atom(v, :D, 2)
        end
    end
    mol
end

d_to_h(mol) = d_to_h!(copy(mol))
h_to_d(mol) = h_to_d!(copy(mol))

Moreover, I have defined matching functions that work independently of the notation. But having a generalised approach that forces a unique represenation is certainly a good idea.

function distinguish_deuterated(mol1::MolGraph, mol2::MolGraph)
    sym1 = atom_symbol(mol1)
    sym2 = atom_symbol(mol2)
    geo1 = hybridization(mol1)
    geo2 = hybridization(mol2)
    isotope1 = Vector{Union{Nothing,Int}}(nothing, length(sym1))
    @inbounds for i in 1:length(sym1)
        if sym1[i] == :D
            sym1[i] = :H
            isotope1[i] = 2
        end
        sym1[i] == :H || continue
        m = MolecularGraph.get_prop(mol1, i, :isotope)
        m ∉ (1, nothing) && (isotope1[i] = m)
    end
    isotope2 = Vector{Union{Nothing,Int}}(nothing, length(sym2))
    @inbounds for i in 1:length(sym2)
        if sym2[i] == :D
            sym2[i] = :H
            isotope2[i] = 2
        end
        sym2[i] == :H || continue
        m = MolecularGraph.get_prop(mol2, i, :isotope)
        m ∉ (1, 0, nothing) && (isotope2[i] = m)
    end

    return (v1, v2) -> sym1[v1] == sym2[v2] && geo1[v1] === geo2[v2] && isotope1[v1] == isotope2[v2]
end

function match_deuterated(mol1::MolGraph, mol2::MolGraph)
    # distinguish D and H
    sym1 = replace(atom_symbol(mol1), :D => :H)
    sym2 = replace(atom_symbol(mol2), :D => :H)
    geo1 = hybridization(mol1)
    geo2 = hybridization(mol2)
    return (v1, v2) -> sym1[v1] == sym2[v2] && geo1[v1] === geo2[v2]
end

[hhaensel]

BTW, what do you think about inchitosdf(), should we replace it to go via inchitomol()?

[mojaie]

Yes, it would be better to replace inchitosdf().
And thank you for letting me know your use case. I'd like to keep calculated descriptors simple by using D => 2H alias, but experimentally introduced VirtualAtom (in ./src/virtualatom.jl) may fit this purpose. This provides custom interfaces for atoms and bonds. For example,

# just example, not implemented yet in the library
struct Deuterium{T<:StandardAtom} <: AbstractAtom
    atom::T
    function Deuterium{T}()
        new(T(; symbol=:H, isotope=2))
    end
end

atom_symbol(atom::Deuterium) = :D

may behave as 2H for calculated descriptors (e.g vmatchgen or standard_weight) without :D => :H mapping, but provide :D to atom_symbol and structure drawing functions that uses atom_symbol.

[hhaensel]

How do we proceed then,? Are you simply merging and adapting inchitosdf or shall I do that?

[mojaie]

Could you add commit for inchitosdf? Anyway I will merge it.

[hhaensel]

Didn't see this, will do ...

[mojaie]

Thank you very much!