| name | genecluster-superpowers |
| description | Use when extending BioSymphony GeneCluster with public-safe sequence or structure search tooling, BGC detection, synteny views, EC prediction, reference databases, interactive viewers, or report rendering. |
GeneCluster Superpowers
This skill collects recommended GeneCluster extension tools from the public survey and test sweep into one invocable kit. Use it as the shortcut for "I want to extend the atlas with X, what's already wired up?"
Single source of truth for tool status: references/docs/biosymphony-tooling-status.md. The status tables in this skill mirror that doc for the originally surveyed subset; the canonical inventory covers the full 25-tool checked set + parked + shelved.
When to use this skill
- Adding a new species to the atlas and you want better-than-BLAST sequence homology (MMseqs2, Foldseek)
- Detecting BGCs that anchor-windowing missed (plantiSMASH 2.0.4 via the checked v7 boot recipe)
- Drawing cross-species cluster synteny ribbons (cblaster + clinker, JCVI MCScan)
- Predicting EC numbers / function for non-Pfam-annotatable candidates (CLEAN-Contact, HIT-EC)
- Layering plant-aware pathway/database context on existing hits (PlantCyc PMN 16, P450Rdb)
- Embedding interactive viewers in atlas reports (Cytoscape.js, copy-paste pattern)
- Rendering the atlas to a publishable HTML book or journal-ready PDF (Quarto, already adopted)
When To Use Something Else
- Running the upstream pipeline itself: use the separate
biosymphony repo skill and the pipeline checkout.
- Routine xlsx postprocess: use the downstream postprocess script from the active project.
- Dispatching provider pipeline jobs: use the active project's runtime checkout. This public skill does not ship runtime launch outputs.
Operating model
Four states per tool: adopted, parked, shelved, or gated.
- Adopted (✅) = checked on RunPod, output integrated into the atlas, canonical path. Just call it.
- Parked (⛔) = install proven on RunPod but a downstream runtime blocker stopped atlas-quality output. Each parked tool has a documented re-entry recipe in
references/docs/biosymphony-tooling-status.md. pick up at the recipe, not at "what's the install path?"
- Shelved (❓) = on the roadmap, license-free, with no dispatch recorded. Listed in the canonical inventory with effort estimate + re-entry recipe; lower priority for current work.
- Gated (⏸️) = blocked on license application or API key access. Alternative checked where available.
Current status (after the extended testing sweep and upstream freshness check, for the full 25-tool inventory see references/docs/biosymphony-tooling-status.md):
| Tool | Status | Tier | Check task | Notes |
|---|
| Quarto 1.9.37 | ✅ ADOPTED | blessed | (adopted) | Quarto atlas book live |
| plantiSMASH 2.0.4 (v7 boot recipe) | ✅ ADOPTED | medium-add | checked | Upstream latest is 2.0.4; "v7" names BioSymphony's checked boot iteration. Raw editable installs had straight.plugin discovery blockers; use the non-editable source install + recipe. Detects multiple clusters per chromosome on public plant genomes. AGPL-3.0+. Review license terms before public service use. |
| antiSMASH 8.0.4 (bacterial / fungal) | ✅ ADOPTED | (added P12) | checked | Cookbook at references/docs/biosymphony-antismash-cookbook.md. Mambaforge + cpu5g. No --taxon plants. |
| DeepBGC (plant BGC default) | ✅ ADOPTED | (added during testing) | checked | Tens of BGCs detected across multiple chromosomes on a public plant genome |
| JCVI MCScan | ✅ ADOPTED | cheap-add | checked | Thousands of pairwise anchors between two public plant species |
| MMseqs2 iterative | ✅ ADOPTED | cheap-add | checked | +8, 21 % homologs vs blastp |
| Foldseek + ProstT5 | ✅ ADOPTED | cross-species support layer | checked | Thousands of PDB structural hits returned on a public plant query set |
| cblaster + clinker | ⛔ PARKED | cheap-add | checked | Re-entry: stage GenBanks via NCBI Datasets CLI -> local cblaster DB -> query. |
| CLEAN + HIT-EC | ⛔ PARKED | cheap-add | checked | Wrapper brittle; re-entry: bypass CLEAN_infer_fasta.py, call CLEAN.infer.infer_maxsep directly. DeepEC / ECPred is the checked EC alternative. |
| HHblits / HHsuite3 | ⛔ PARKED | optional | checked | cpu3g OOM on PDB70; re-entry: cpu5g + pre-stage PDB70 in separate pod. MMseqs2 + Foldseek cover divergent-homolog space. |
| PlantCyc PMN 16 | ⏸️ GATED | DB-only | . | Academic license required. KEGG / KAAS covers pathway-completion in the interim. |
| P450Rdb | ✅ ADOPTED | DB-only | checked | 3 BIA queries at 100 % identity |
| Cytoscape.js 3.33.3 | ✅ ADOPTED | frontend | . | Pathway-completion viewer with Fit/Reset/zoom controls |
Also checked in the extended test campaigns: MIBiG 4.0, ESM-C 300M Synthyra, ESM-2 650M, HMMER, InterProScan, TM-align / mTM-align, MAFFT, IQ-TREE, ColabFold, KEGG mapper / KAAS, EnzymeMap, DiffPaSS, DeepEC / ECPred, igv-reports, pyGenomeTracks, and Quarto Dashboards.
The canonical roadmap with original verdict matrices and citations: references/docs/biosymphony-genecluster-superpower-roadmap.md (pre-test predictions, current state lives in references/docs/biosymphony-tooling-status.md).
The per-tool integration plans with install + sample CLI: references/docs/tooling/<tool>.md.
The per-tool quickstarts (terse + actionable): references/<tool>-quickstart.md (in this skill).
Quick start: what's installed?
Before you do anything, check what's available:
bash scripts/superpowers-status.sh
Output shows which tools are installed, their version, and the install command for the missing ones.
Install tiers
Three idempotent install scripts at tools/recommended/:
bash tools/recommended/install-cheap.sh
bash tools/recommended/install-medium.sh
bash tools/recommended/install-heavy.sh
You usually want cheap-tier first; medium and heavy land when you need the specific capability.
Per-tool quickstarts
Each references/<tool>-quickstart.md has the same shape: status, install one-liner, sample run on atlas data, integration in our pipeline, open questions, see-also links. The runner scripts in scripts/run-<tool>.sh mirror the sample-run command with sane defaults pointing at .runtime/campaign-<species>-summary/.
Adopted (just use):
Parked. re-entry recipe required before retry (read references/docs/biosymphony-tooling-status.md first):
Sample invocation pattern
Every runner takes a species shorthand as $1:
bash scripts/run-cblaster.sh <species>
bash scripts/run-jcvi-mcscan.sh <species-a> <species-b>
bash scripts/run-mmseqs2.sh <species>
bash scripts/run-foldseek-prostt5.sh <species>
If the underlying tool is not installed, each runner exits cleanly with the exact install command.
Required checks before using this skill
bash scripts/superpowers-status.sh
ATLAS_SUMMARY_DIR=/path/to/your/atlas-summary
test -d "$ATLAS_SUMMARY_DIR"
QUARTO_PROJECT=/path/to/your/quarto-atlas
test -f "$QUARTO_PROJECT/_quarto.yml" && (cd "$QUARTO_PROJECT" && quarto render --to html)
Run these any time before kicking off a tool integration.
When to add a new tool to this skill
When a new tool gets recommended (via a focused tooling review or a new paper), the integration follows the foundation-laying pattern:
- Survey & decide, usually via a focused tooling review; end with a verdict in the superpower roadmap doc.
- Doc, add
references/docs/tooling/<tool>.md with install + sample CLI + integration point + open questions.
- Install script, extend
tools/recommended/install-{cheap,medium,heavy}.sh with the new install (pin version where stable, leave a "or latest" comment otherwise).
- Runner, add
tools/recommended/<tool>/run-on-species.sh.template for the canonical CLI shape, then promote to scripts/run-<tool>.sh once the runner is concrete.
- Quickstart, add
references/<tool>-quickstart.md (300 words or fewer, terse).
- Status table, update this
SKILL.md's status table + the tools/recommended/README.md status table + references/docs/biosymphony-genecluster-superpower-roadmap.md.
- Lessons, if the tool's adoption codifies a non-obvious lesson (Cytoscape gradient bug, Quarto stub references, etc.), capture it inline in the quickstart and status table so future readers see it without external context.
When a tool gets ADOPTED (installed, integrated, used in production), promote its row in the status table from "foundation" to "✓ ADOPTED" and update its quickstart's status header. If upstream has cut a newer release, record both the upstream release and the BioSymphony checked pin until a smoke test proves the upgrade.
Cross-references
- Tool roadmap with verdict matrices:
references/docs/biosymphony-genecluster-superpower-roadmap.md
- Per-tool integration docs:
references/docs/tooling/<tool>.md (10 files)
- Atlas authoring best practices:
references/docs/README.md
- Quarto rendering runbook:
references/docs/README.md
- Obsidian view companion:
references/docs/biosymphony-atlas-obsidian-walkthrough.md
- Pipeline / RunPod dispatch: the
biosymphony repo skill plus the pipeline checkout
Provenance
Discovery survey: parallel tool surveys along five angles (sequence/structure search, BGC detection, enzyme function, reporting, comparative-atlas paper styling). Cross-checking picked five high-confidence adds (Quarto adopted; plantiSMASH + cblaster + JCVI + Foldseek + CLEAN queued for adoption when use case lands).
Test sweep: Public-safe provider-side checks covered every tool above plus 17 extended tools. Result: 25 checked, 3 parked (cblaster + clinker, CLEAN + HIT-EC, HHblits), 2 gated (PlantCyc, ESM-C 6B). Raw run logs are omitted from this public snapshot; see references/docs/biosymphony-tooling-status.md for the canonical public inventory and re-entry recipes.
Freshness check: Verified current release endpoints for the explicitly pinned tools. Corrections applied here: plantiSMASH is upstream 2.0.4 with a BioSymphony v7 boot recipe rather than upstream "v7"; antiSMASH current line is 8.0.4; local-only pins for clinker, JCVI, igv-reports, and Cytoscape.js were refreshed.
Cloud-runtime forward research: 2026-05-11. Provider-portability reference docs for AWS / GCP / neocloud live at references/docs/cloud-runtimes/. Result: RunPod stays default for both CPU and GPU; clear exceptions documented (NCBI-bandwidth-bound work -> GCP, formal-review work -> AWS, large-scale H100 training -> CoreWeave).
This skill and the linked inventory are durable shortcuts. Future iterations can consult this skill, read the inventory, and pick the right starting point.