Module end to end

Module: end-to-end

Runs annotate, then de-novo (reusing the proteins predicted by annotate), then aggregate, writing each module to its own sub-directory within a single output:

<output>/
|-- 01_annotate/
|-- 02_denovo/
`-- 03_aggregate/

This is the recommended way to run CheckAMG when feasible. Because de-novo is more sensitive than annotate alone, the combined run often identifies additional AVGs that annotation cannot, while still providing the functional classifications and curation from annotate. de-novo benefits substantially from a GPU; if a GPU is unavailable or de-novo is impractical at your input scale, run annotate alone instead.

When --input-type nucl, the de-novo step reuses the proteins predicted by the annotate module (proteins are not predicted twice). The annotate step always runs on CPU; de-novo can be directed to a GPU independently via --denovo-accelerator.

Basic usage

checkamg end-to-end \
  -d /path/to/db/destination \
  -i examples/example_data/single_contig_viruses.fasta \
  -I examples/example_data/multi_contig_vMAGs \
  -o CheckAMG_example_out

--db-dir must point to a directory containing both the annotate and de-novo databases (the default layout produced by checkamg download).

Parameters

Required

Argument	Description
-d, --db-dir	Directory containing both the annotate and de-novo databases.
-o, --output	Output directory for all modules and combined results.

Inputs

Same as annotate: -i/--input-contigs, -I/--input-bins, -p/--input-proteins, -P/--input-bin-proteins, and --input-type (nucl or prot, default nucl).

annotate options

A curated subset of the annotate parameters is exposed (the rest use annotate's defaults):

Argument	Default	Description
-l, --min-len	5000	Minimum contig length (bp).
-f, --min-orf	4	Minimum ORFs/proteins per contig.
-amg, --min-amg-weight	0.6	Minimum AMG weight for a putative AMG annotation to be kept.
-kf, --keep-full-hmm-results / --no-...	off	Write all HMM search results (large; use with --save-to-parquet).
--filter-ambig-regions / --no-...	off	Exclude predictions outside strict viral regions.
--filter-avg-arrays / --no-...	on	Exclude AVG predictions in contiguous arrays.
--avg-array-limit	3	Exclude AVG runs of this length or more.
--filter-presets	allow_glucan,allow_nucleotide,allow_methyl,allow_lipid	Annotation filtering presets (see annotate).

de-novo options

Argument	Default	Description
-b, --batch-size	16	Batch size in scaffolds when computing PST embeddings.
--nn-batch-size	4096	Batch size in proteins for kNN searches.
--num-index-cells	auto	Cells to partition the training data into when building a search index.
--num-probe-cells	1	Index cells to visit when searching for nearest neighbors.
-k, --knn	20	Number of nearest neighbors for distance-weighted voting.
--denovo-accelerator	auto	Accelerator for de-novo only (cpu/gpu/auto). Annotate always runs on CPU.
--denovo-devices	= --threads (CPU)	Number of devices for de-novo. On CPU this is cores/threads; capped at 1 on GPU.

Outputs, resources

Argument	Default	Description
-pq, --save-to-parquet / --no-...	off	Write tables as parquet instead of TSV.
-t, --threads	25% of available	Maximum number of threads.
-m, --mem	80% of available	Memory limit in GB.
--debug / --no-debug	off	Enable debug-level logging.

Outputs

• 01_annotate/: a full annotate output.
• 02_denovo/: a full de-novo output.
• 03_aggregate/: the merged aggregate report (aggregated_results.tsv, aggregated_results_detailed.tsv, and aggregated_results_categories.tsv). The categories table can list multiple function categories per AVG (especially AMGs); this does not mean the gene acts in all of them (or any). See Why does one AVG have multiple function categories?.