Preprocessing

MolecularGraph.jl version: 0.17.1

This tutorial includes following preprocessing strategies.

Remove hydrogen vertices
Extract molecules of interest
Standardize charges
Dealing with resonance structure
Customize property updater

👀 Reading hidden code

45.3 ms

Public databases (e.g. PubChem, ChEMBL) and flat file databases (e.g. SDFile) have different formats and may not always be used for your analysis as is. For example,

whether hydrogens are explicitly written or omitted
whether salt and water molecules are included in the molecular graph or provided as metadata
representation of resonance structure (e.g. diazo group; [C-]-[N+]#N <-> C-[N+]=[N-])
charges depend on the condition - powder, dissolved or in physiological condition

👀 Reading hidden code

184 ms

using Graphs, MolecularGraph

👀 Reading hidden code

3.8 s

"_data"

data_dir = let

# Create data directory

data_dir = "_data"

isdir(data_dir) || mkdir(data_dir)

data_dir

end

👀 Reading hidden code

79.9 μs

fetch_mol! (generic function with 1 method)

function fetch_mol!(cid, name, datadir)

url = "https://pubchem.ncbi.nlm.nih.gov/rest/pug/compound/cid/$(cid)/SDF"

dest = joinpath(data_dir, "$(name).mol")

isfile(dest) || download(url, dest);

return dest

end

👀 Reading hidden code

1.2 ms

"_data/Cefditoren Pivoxil.mol"

molfile = fetch_mol!("6437877", "Cefditoren Pivoxil", data_dir)

👀 Reading hidden code

56.2 μs

Remove hydrogen vertices

SDFiles downloaded from PubChem have hydrogen nodes. In practice, hydrogens which is not important are removed from molecular graphs for simplicity.
remove_hydrogens!(mol) removes hydrogen vertices that are not important (no charge, no unpaired electron, no specific isotope composition and not involved in stereochemistry).
remove_all_hydrogens!(mol) removes all hydrogen vertices.

👀 Reading hidden code

116 ms

mol = sdftomol(molfile)

👀 Reading hidden code

13.2 s

let

remove_hydrogens!(mol)

mol

end

👀 Reading hidden code

358 ms

Extract molecules of interest

connected_components(mol) returns connected components that are sets of vertices of the individual molecules in the molecular graph object.

👀 Reading hidden code

665 μs

mol2 = smilestomol("CC(=O)OC1=CC=CC=C1C(=O)O.O.CCO"); nothing

👀 Reading hidden code

6.5 s

html_fixed_size(mol2, 250, 250, atomindex=true)

👀 Reading hidden code

1.2 s

Vector{Int64}

Int64

connected_components(mol2)

👀 Reading hidden code

36.9 μs

To extract the molecule of interest, you can iterate over the connected components and apply induced_subgraph(mol, vertices) to extract the molecules and filter them one by one.
Or simply extract_largest_component!(mol) can be used. This removes vertices not belong to the largest component (connected component which has the largest number of vertices) from the graph.

👀 Reading hidden code

798 μs

let

mol = copy(mol2)

extract_largest_component!(mol)

mol

end

👀 Reading hidden code

1.0 s

Standardize charges

protonate_acids!(mol) removes charges on oxo/thio acid anions
deprotonate_oniums!(mol) removes charges on ammonium/oxonium cations

👀 Reading hidden code

747 μs

charged = smilestomol("CCCC(=O)[O-].[N+]CCCC[N+]")

👀 Reading hidden code

624 μs

let

mol = copy(charged)

protonate_acids!(mol)

deprotonate_oniums!(mol)

mol

end

👀 Reading hidden code

242 ms

Dealing with resonance structure

Substructure match methods in this library compares atom symbols and the number of $\pi$ electrons, so in many cases you don't have to care about fluctuations in resonance structure.

👀 Reading hidden code

9.2 ms

quinoline1 = smilestomol("N1=CC=CC2=C1C=CC=C2")

👀 Reading hidden code

437 μs

quinoline2 = smilestomol("N=1C=CC=C2C1C=CC=C2")

👀 Reading hidden code

643 μs

Loading more cells...

In the cloud (experimental)

On your computer

Frontmatter

Preview

Preprocessing

Remove hydrogen vertices

Extract molecules of interest

Standardize charges

Dealing with resonance structure