Run Perturb-seq direct-guide capture with a local agent
This route assigns a reviewed fixed read window to a finite direct-guide list. It is the same deterministic assignment contract used by CRISPR counting, with the narrower scientific boundary required for Perturb-seq guide-capture reads.
Exact inputs
targets: local TSV or CSV containing the known guide-barcode identifiers and sequences;reads_dir: local directory containing only the intended guide-capture FASTQ or FASTQ.gz files;output_dir: absent or empty local directory;optional
threads,max_reads, andmax_startintegers.
The target table must be known before assignment. DotMatch does not discover new guide sequences from the reads.
Copyable start
Create perturb-seq-request.json:
{
"intent": "perturb-seq-guide-capture",
"targets": "/absolute/path/direct-guides.tsv",
"reads_dir": "/absolute/path/guide-capture-fastqs",
"output_dir": "/absolute/path/dotmatch-perturb-seq-run",
"threads": 4
}
dotmatch agent invoke prepare_assay --input perturb-seq-request.json
Pass the returned spec through inspect_assay, run_assay,
review_assay, and—only when the reliability boundary permits it—
handoff_assay. The JSON shapes are identical to the
CRISPR agent route.
Outputs
inferred AssaySpec and inference evidence;
per-guide unique counts and optional per-read assignments;
separate ambiguous, unmatched, and invalid outcomes in QC and findings;
reliability, provenance, artifact hashes, resource use, and raw-data-free handoff records.
Evidence boundary
The GSE146194 reference evaluates a frozen 32-guide direct-capture rule on
48,000 held-out reads. DotMatch and matched independent per-read oracles have
zero differences at k=0 and k=1 for that recorded dataset, target list,
window, orientation, metric, and ambiguity rule.
This does not establish cell-barcode correction, UMI deduplication, guide-per-cell calls, expression quantification, indirect-capture designs, perturbation effects, or biological conclusions. Those steps and claims must remain outside the DotMatch handoff.