SynbioSynthetic Biology

Genetic Circuit Design API: Cello and CRISPR Guide RNA Scoring

Cello genetic CAD, SynBioHub parts registry, DeepCRISPR guide scoring — all fragmented. SciRouter brings design-build-test-learn under one API.

SciRouter Team
April 11, 2026
13 min read

Synthetic biology is the engineering discipline that treats DNA as code. You specify what you want a cell to do; a set of tools compiles that specification into a DNA sequence; a lab assembles the sequence and transforms it into a host; measurements feed back into the next design round. This design-build-test-learn (DBTL) loop has been the central metaphor of the field for over a decade, but the tools required to run it at cloud speed were scattered across a dozen packages with incompatible interfaces.

SciRouter's synbio API unifies them. You can compile a genetic circuit with Cello, query parts from SynBioHub, score CRISPR guide RNAs with DeepCRISPR, and optimize codon usage for your host organism, all from a single API. It is the closest thing to a compiler toolchain that synthetic biology has.

Note
The API includes automated biosecurity screening against public databases of select agents and regulated sequences. Designs that match high-risk targets are flagged for manual review. Users remain responsible for compliance with institutional biosafety and applicable regulations.

Cello: compiler for genetic circuits

Cello started at MIT as a tool for automating E. coli genetic circuit design. You write a Verilog-style specification (truth table, logic function, or a small language program), pick a target organism from a supported list, and Cello returns a full DNA sequence that implements your specification. Under the hood it uses a characterized library of parts (promoters, ribosome binding sites, terminators, logic gates) and a technology mapping step borrowed from electronic design automation.

Cello is one of the most cited tools in synbio for a reason. It transformed circuit design from a manual exercise in part selection into something that looks like compiling a program. Open source releases have expanded it to yeast and to more complex logic.

  • Input. Truth table or Verilog-like specification.
  • Output. Full DNA sequence in SBOL or GenBank format, annotated with each part's role.
  • Supported organisms. E. coli (native), yeast (via extensions), with community contributions for other hosts.

SynBioHub: the parts registry

Behind every circuit design is a catalog of biological parts: promoters with measured strengths, ribosome binding sites with known translation efficiencies, reporters that are well characterized in your host. SynBioHub is the open source repository where the community curates these parts using the SBOL standard. As of 2026 it contains tens of thousands of parts across several host organisms.

The API lets you query SynBioHub directly. Search by function (give me all strong constitutive promoters for yeast), by family (all variants of GFP), or by full text. Results come back as SBOL documents that you can drop straight into a Cello design call.

A circuit design call

design-circuit.py
import httpx

API_KEY = "sk-sci-..."
BASE = "https://scirouter.ai/v1"

# Design a simple two-input AND gate in E. coli
response = httpx.post(
    f"{BASE}/synbio/design-circuit",
    headers={"Authorization": f"Bearer {API_KEY}"},
    json={
        "tool": "cello",
        "organism": "e-coli",
        "specification": {
            "inputs": ["aTc", "IPTG"],
            "outputs": ["YFP"],
            "truth_table": {
                "00": 0,
                "01": 0,
                "10": 0,
                "11": 1,
            },
        },
        "parts_library": "cello-eco1c1g1t1",
        "optimize": "expression",
    },
    timeout=300,
)

result = response.json()
print(f"Design score: {result['score']:.2f}")
print(f"Sequence length: {len(result['sequence'])} bp")
print(f"GenBank URL: {result['genbank_url']}")

The call returns a ready-to-order DNA sequence with each part annotated, a predicted performance score, and a GenBank file suitable for gene synthesis. The biosecurity screen runs automatically before the response is returned.

CRISPR guide RNA scoring

CRISPR is the other pillar of modern synthetic biology, and guide RNA design is where most of the machine learning action happens. A good guide binds its target efficiently and does not hit off-targets elsewhere in the genome. Predicting both from sequence alone is a supervised learning problem with abundant training data, and models like DeepCRISPR, CRISPRon, and DeepSpCas9 have pushed accuracy well past the level of classical scoring rules.

The API exposes three complementary capabilities:

  • On-target efficiency prediction. How well will this guide cut its target? Scored in the 0 to 1 range with calibration against experimental datasets.
  • Off-target risk scoring. For each candidate guide, scan the reference genome and return a ranked list of potential off-targets with CFD-style scores.
  • Guide library generation. Given a gene list and a host genome, generate a full guide library with on- and off-target scores, ranked for selection.

Which scoring model to use

Different models are trained on different cell types and Cas9 variants. The recommended default is DeepSpCas9 for standard SpCas9 in mammalian cells, CRISPRon for knockout screens, and a specialized model for base editors or prime editing. The API exposes all of them and lets you ensemble across models for additional robustness.

Codon optimization

Every new heterologous expression experiment requires codon optimization to match the target host. The API includes codon optimization backends (CAI-based, GC content-targeted, and machine learning models trained on expression datasets). You supply a protein sequence and a host, and get back a codon- optimized DNA that should express well.

Codon optimization is also one of the places where safety matters. A newly generated sequence should be screened for unintended reading frames, cryptic promoters, and resemblance to regulated agents. The API runs these checks automatically.

Design-build-test-learn as one API

The real power of a unified synbio API is running the whole DBTL loop from one place. A typical workflow:

  • Design. Use Cello or a custom compiler to generate candidate circuits.
  • Build. Codon-optimize, design assembly primers, and generate order-ready DNA sequences.
  • Test. For each candidate, predict expression levels, protein solubility (SoluProt via the protein API), and stability (ThermoMPNN). Rank candidates before synthesis.
  • Learn. Feed measured results back into active learning models that propose the next round of designs.

Each step is a single API call and the intermediate artifacts (SBOL documents, sequences, score tables) flow between them without reformatting. That is what makes the loop practical to run at cloud speed.

Tip
For heterologous expression projects, running the protein API (solubility, stability) on your codon-optimized sequence before ordering DNA is cheap insurance. A 10-minute in silico check can save weeks of wet-lab troubleshooting on a poorly expressed construct.

Who is using the synbio API

  • Academic labs running iGEM teams and teaching synthetic biology courses.
  • Biotech startups designing expression constructs for new therapeutic targets.
  • Bio-manufacturing companies engineering strains for chemicals and materials.
  • CRISPR screen designers generating genome-scale guide libraries.

Biosecurity and responsible use

Biosecurity screening is non-negotiable for a hosted synbio API. The service runs every designed sequence against public databases of regulated agents and known virulence factors. Matches above a threshold are blocked or flagged for human review. This is the same kind of screening responsible gene synthesis providers perform before shipping DNA.

Users remain responsible for compliance with their institution's biosafety committee, local regulations, and international conventions on dual-use research. The API provides the tools; users provide the judgment.

Getting started

Start with Synbio Lab, the web interface for circuit design, parts search, and guide RNA scoring. It supports truth-table-based circuit design in the browser without requiring a local install.

For production pipelines, the Python SDK gives you the full API surface plus convenient wrappers for common design tasks. Research teams building multi-step DBTL workflows typically orchestrate them from Python or from an AI agent using the MCP server.

Open Synbio Lab →

Frequently Asked Questions

What is Cello and why is it important for synthetic biology?

Cello is a genetic circuit design tool developed at MIT. You write a truth table describing the logic you want a cell to compute, and Cello returns a DNA sequence that implements that logic using a library of characterized biological parts (promoters, RBSs, terminators, logic gates). It brought compiler-style automation to genetic engineering and is one of the most cited tools in synthetic biology. The API wraps the open source version so you do not have to install it locally.

What is SynBioHub?

SynBioHub is an open source repository and search engine for biological parts based on the SBOL (Synthetic Biology Open Language) standard. It hosts tens of thousands of characterized promoters, coding sequences, terminators, and whole circuits contributed by the research community. The API queries SynBioHub programmatically so you can pull parts directly into a design workflow without manual copy-paste.

What is DeepCRISPR and how accurate are CRISPR guide scoring models?

DeepCRISPR is one of several machine learning models that predict CRISPR guide RNA efficiency and off-target risk from sequence alone. Recent models including DeepSpCas9, CRISPRon, and DeepCRISPR v2 achieve correlations above 0.85 with measured activity on held-out datasets. They are reliable for ranking candidate guides but still benefit from experimental validation before committing to a large screen.

Can I design a full design-build-test-learn cycle through the API?

Yes, the API is intentionally structured around the DBTL loop. Design endpoints handle circuit design and sequence optimization. Build endpoints help with assembly planning and primer design. Test endpoints score candidate variants (expression, solubility, stability) in silico. Learn endpoints feed measured results back into active learning models. A full loop can be orchestrated from a single script.

What host organisms are supported?

The Cello wrappers support E. coli (the original target) and yeast. Part libraries exist for Bacillus subtilis, Pseudomonas putida, and a handful of mammalian systems. For CRISPR guide scoring, the major models are host-agnostic because they focus on the guide-target interaction, but you should always check that the scoring model was trained on data relevant to your cell type.

How does the API handle safety and biosecurity?

The API screens designed sequences against public databases of select agents, regulated toxins, and known virulence factors before returning them. Designs that match high-risk sequences are flagged and require manual review. This is the same kind of screening that responsible gene synthesis providers perform. Users are still responsible for complying with their institution's biosafety requirements and applicable regulations.

Run this yourself — no GPU, no install

Free for researchers. Pick a tool, paste your input, see results in seconds.