af3-json-sha256

name	af3-json-sha256
description	Extract deduplicated protein, DNA, and RNA sequences from AF3 or Protenix inference JSON files into per-entity-type FASTA files keyed by SHA-256 hash. Use as the first stage of the MSA pipeline before `af3-msa-search` and `af3-msa-injector`.

AF3/Protenix JSON to FASTA (SHA-256 keyed)

Extract unique biological entity sequences from AF3 or Protenix inference JSON and write one FASTA file per entity type (protein, DNA, RNA), with each entry keyed by its SHA-256 hash.

Pipeline Context

This is stage 1 of the structure prediction MSA pipeline:

1. af3-json-sha256 (this skill) — JSON to deduplicated FASTA
2. af3-msa-search — ColabFold MSA search from FASTA
3. af3-msa-injector — Inject MSA results back into inference JSON

Run commands from repository root and prefer .venv/bin/python.

Quick Start

1. Extract from a single JSON file:

.venv/bin/python skills/af3-json-sha256/scripts/af3_json_to_fasta.py \
  --input /abs/path/to/job_xxx.json \
  --output /abs/path/to/fasta_out

2. Extract from a directory of JSON files:

.venv/bin/python skills/af3-json-sha256/scripts/af3_json_to_fasta.py \
  --input /abs/path/to/json_dir \
  --output /abs/path/to/fasta_out \
  --glob '*.json'

3. Extract only protein sequences:

.venv/bin/python skills/af3-json-sha256/scripts/af3_json_to_fasta.py \
  --input /abs/path/to/json_dir \
  --output /abs/path/to/fasta_out \
  --types protein

Output Format

One FASTA file per entity type is written to --output:

• protein.fasta — all unique protein sequences
• dna.fasta — all unique DNA sequences
• rna.fasta — all unique RNA sequences

Empty entity types produce no file. FASTA entries are sorted by hash for deterministic output.

>a1b2c3d4e5f6...  (64-char SHA-256 hex digest)
NORMALIZEDSEQUENCE

Sequences are wrapped at 60 characters per line.

CLI Flags

• --input: Single JSON file or directory of JSON files (required).
• --output: Output directory for FASTA files (required).
• --format auto|af3|protenix: Force format or auto-detect (default: auto).
• --glob: File pattern when --input is a directory (default: *.json).
• --types: Comma-separated entity types to extract (default: protein,dna,rna).

Entity Type Mapping

AF3 key	Protenix key	Entity type	Extracted?
protein	proteinChain	protein	Yes
dna	dnaSequence	dna	Yes
rna	rnaSequence	rna	Yes
ligand	ligand	—	No
ion	ion	—	No

Sequence Normalization

All sequences are normalized before hashing:

1. Strip all whitespace
2. Convert to uppercase

The same normalization is used by af3-msa-injector for SHA-256 matching, ensuring hash consistency across the pipeline.

Resources

• scripts/af3_json_to_fasta.py: deterministic CLI for FASTA extraction from AF3/Protenix JSON.