Well, turned out that
"Fixed a bug that caused RDF/XML long output...", mentioned in the Paxtools v5.2.1,v5.3.0 release anouncement last year, actually intruduced an older regression bug: rdf:datatype = "xsd:string" attribute inside e.g. bp:name, bp:db, etc. biopax properties in RDFXML does not make sense or is just wrong (affects some SPARQL and RDF tool/converter use cases); so paxtools-core i/o (SimpleIOHandler) either should not add rdf:datatype when serializing a BioPAX Model to RDFXML (because biopax-level3.owl defines each property's range anyway, and RDF tools should be able to find out, infer that) or use absolute IRI there like before, e.g., rdf:datatype = "
http://www.w3.org/2001/XMLSchema#string", or could use XML entity, e.g. rdf:datatype = "&xsd;string";
see #54)
This is going to be fixed in Paxtools v6.0.0 and may be v5.2.2 as well.