generate$from.pgs.file

generate$from.pgs.file() reads PGS Catalog scoring files from disk into PolyGenius. It is the local-file counterpart to generate$from.pgs.catalog(), which imports the same format over the network by accession.

A PGS when path is one file; a PGSLibrary when it is a directory or a vector of paths. A returned set carries a record with name = "generate$from.pgs.file", the deparsed call, and a sources entry recording the path read; see the Library-level provenance section of PGSLibrary. A single model has no library-level record -- its own $generation names the scoring file it came from.

Files written by PGS$write.pgs.file() in extended mode round-trip losslessly, because their #polygenius_* extension keys restore gwas, generation, liftover and any user fields. Plain PGS Catalog or third-party files load as leaner models built from the standard header alone.

A directory containing a manifest.json (written by PGSLibrary$write.pgs.file()) is read in the manifest's order, and the manifest is verified twice: its position values must form a gap-free 1..M permutation, and the fingerprint of the set just read must equal the recorded one. Either check failing aborts, so a file added, removed or edited since the manifest was written is not silently accepted. A directory without a manifest is read as every *.txt/*.txt.gz it holds, in list.files() order.

The same two guarantees as generate$from.pgs.catalog apply, because both paths share one import boundary. $variants$beta is converted to a log-scale per-allele effect according to the file's #weight_type header; an unreported scale warns, an uninterpretable one is refused, and a file whose #genome_build is absent, NR, or an unsupported assembly is refused. $build holds the canonical name ("hg19/GRCh37" or "hg38/GRCh38") whichever alias the file declares, and a written file declares GRCh37 or GRCh38.

Rows repeating a variant (same chromosome, position, other and effect allele) load once: exact copies keep the first row, and a variant listed with different weights is dropped. Either case warns. See PGS().

A file written by PGS$write.pgs.file() always declares #weight_type=beta, because the conversion already happened when the model was imported, so a write/read round-trip does not convert a second time. The scale the score was originally published on survives in the #polygenius_generation extension key.

The model is named after the file's #pgs_name header, falling back to its #pgs_id accession when no usable name is declared. A file carrying neither names no score and is refused: the path a file happens to sit at is not a name the file gives itself.