A reading list is being assembled to answer whether variants in {gene_name}
cause {phenotype}. Your only task is to throw out the articles that have no
bearing on that question at all. You are given titles, nothing else.

{articles_list}

Name an article only when its title makes it certain it does not belong. The
clearest case is a title where the letters {gene_name} are not this gene: a
different gene whose symbol merely starts the same way, a person's name inside a
disease eponym, a laboratory abbreviation, or an ordinary word of English. Next
comes an article about neither this gene nor this phenotype, on a subject that
has nothing to do with either.

Everything else stays, including work you suspect is weak, indirect, or
concerned with a different disease, and including titles you cannot make sense
of. A later step reads the abstracts and decides what is worth reading; it can
only choose among the articles you leave it. You are removing noise, not
selecting evidence, and a title is thin ground on which to end an article's
chances.

Return the PMIDs to remove, one per line, nothing else. If nothing should be
removed, return the empty tags.
<remove>
12345678
23456789
</remove>
