Incorrect gene reaction rule parsing

[stelmo]

Currently if genes are not in DNF format they are not parsed correctly. E.g. if a model has:

"gene_reaction_rule":"Mbar_A0384 and Mbar_A0385 and Mbar_A0389 and (Mbar_A0391 or Mbar_A0379) and (Mbar_A3102 or Mbar_A0386) and (Mbar_A0390 or Mbar_A0378) and Mbar_A0387 and Mbar_A0392 and (Mbar_A3101 or Mbar_A0388)",

then this is yielded by the accessor:

julia> reaction_gene_association(model, "ATPS4r")
1-element Vector{Vector{String}}:
 ["Mbar_A0384", "Mbar_A0389", "Mbar_A0387", "Mbar_A0392", "Mbar_A0385"]

From IAF692.json on BIGG

[exaexa]

OK anyway, this is one possible parser:

import PikaParser as P

abstract type LogicalExpr end

struct AndExpr <: LogicalExpr
    l::LogicalExpr
    r::LogicalExpr
end

struct OrExpr <: LogicalExpr
    l::LogicalExpr
    r::LogicalExpr
end

struct NotExpr <: LogicalExpr
    x::LogicalExpr
end

struct IdentExpr <: LogicalExpr
    id::Symbol
end

isident(x::Char) = isletter(x) || isdigit(x) || x == '_' || x == '-' || x == ':' || x == '.'

noglue(x) = P.seq(x, P.not_followed_by(:identpart))

rules = Dict(
    :identpart => P.satisfy(isident),
    :space => P.many(P.satisfy(isspace)),
    :id => noglue(:in_id => P.some(:identpart)),
    :orop => P.first(P.tokens("||"), P.token('|'), noglue(P.tokens("OR")), noglue(P.tokens("or"))),
    :andop => P.first(P.tokens("&&"), P.token('&'), noglue(P.tokens("AND")), noglue(P.tokens("and"))),
    :notop => P.first(P.token('!'), noglue(P.tokens("NOT")), noglue(P.tokens("not"))),
    :expr => P.seq(:space, :orexpr, :space),
    :orexpr => P.first(
        :or => P.seq(:andexpr, :space, :orop, :space, :orexpr),
        :andexpr,
    ),
    :andexpr => P.first(
        :and => P.seq(:notexpr, :space, :andop, :space, :andexpr),
        :notexpr,
    ),
    :notexpr => P.first(:not => P.seq(:notop, :notexpr), :baseexpr),
    :baseexpr => P.first(:id, :parenexpr => P.seq(P.token('('), :space, :orexpr, :space, P.token(')'))),
)

grmr = P.make_grammar([:expr], P.flatten(rules, Char))

input = " Mbar_A0384 and Mbar_A0385 and Mbar_A0389 and (Mbar_A0391 or Mbar_A0379) and (Mbar_A3102 or Mbar_A0386) and (Mbar_A0390 or Mbar_A0378) and Mbar_A0387 and Mbar_A0392 and (Mbar_A3101 or Mbar_A0388) "

s = P.parse(grmr, input)

m = P.find_match_at!(s, :expr, 1)
@assert s.matches[m].len == length(input) "match not found"

openfn(m,_) = m.rule == :expr ? Bool[0,1,0] :
        m.rule == :parenexpr ? Bool[0,0,1,0,0] :
        m.rule == :not ? Bool[0,1] :
        m.rule in [:or, :and] ? Bool[1,0,0,0,1] :
        m.rule in [:andexpr, :orexpr, :notexpr, :baseexpr] ? Bool[1] :
        (false for _ in m.submatches)

foldfn(m,_,subvals) = m.rule == :id ? IdentExpr(Symbol(m.view)) :
        m.rule == :not ? NotExpr(subvals[2]) :
        m.rule == :and ? AndExpr(subvals[1], subvals[5]) :
        m.rule == :or ? OrExpr(subvals[1], subvals[5]) :
        m.rule == :parenexpr ? subvals[3] :
        m.rule == :expr ? subvals[2] :
        isempty(subvals) ? nothing : subvals[1]

grr = P.traverse_match(s, m, open = openfn, fold = foldfn)

Questions arise:

• Do we support negation? (in DNF, we'd need to have slightly more complicated clauses)
• How come we didn't need negation yet? is it just so uncommon in metabolic modeling?

[exaexa]

(the @assert there will get simplified once I get time to implement this https://github.com/LCSB-BioCore/PikaParser.jl/issues/15 )

[exaexa]

EDIT: we don't need not because apparently no one actually uses that, the gene logic is plain constructive.

[stelmo]

I just checked, this works! nice!

[exaexa]

Additional fun:

• Fruitfly GEM uses colons in gene names
• there may be unicode in gene names (but that's hopefully already handled by the parser above)

[exaexa]

also:

• square brackets
• dots
• matt's gene

[stelmo]

this was addressed by #731