Cache Handling¶

phu caches predicted proteins for both screen and jack so repeated runs can skip the gene-prediction step when the contigs and prediction inputs are unchanged.

What the cache stores¶

The cache stores the translated protein FASTA generated from the input contigs. Search-specific inputs such as HMM files, seed markers, combine mode, and output folder are not part of the cache key.

When the cache is reused¶

The cached proteins are reused when all prediction inputs are the same:

the input contigs file has not changed
the prediction mode is unchanged
the protein-length filter is unchanged
the gene-length filter is unchanged
the translation table is unchanged

When the cache is rebuilt¶

The cache is invalidated and rebuilt when any of these prediction inputs change:

--mode
--min-gene-len in phu jack
--min-protein-len-aa
--ttable

For phu screen, changing --min-protein-len-aa, --mode, or --ttable forces a rebuild. For phu jack, changing either --min-gene-len or --min-protein-len-aa forces a rebuild because both values feed the shared protein-prediction cache key.

How to control it¶

Cache behavior is controlled by environment variables:

PHU_CACHE=off disables caching and always recomputes proteins.
PHU_CACHE_DIR overrides the cache location.
If PHU_CACHE_DIR is not set, phu uses an XDG-style cache directory under your home cache path.

You can also remove all cached predictions explicitly:

phu --clean-cache

This removes the full cache directory resolved by PHU_CACHE_DIR (or the default cache path) and exits.

Practical cases¶

Changing only HMM files in screen reuses the cached proteins.
Changing only seed markers in jack reuses the cached proteins.
Changing the protein-length filter rebuilds the cache, which is expected because it changes the predicted protein set.