Industries / Frontier science / Synthetic biology
Design tools
This work is the software, standards and shared data used to specify a biological design, from a single part to a pathway, before anything is made.
7
Milestones achieved, from 2000 to 2021.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2000
A metabolism as a matrix
The annotated genome of one E. coli strain was turned into a table of every metabolic reaction and the amounts it consumes and produces, which let the strain's metabolic capabilities be computed rather than only measured.
Credit
Jeremy S. Edwards and Bernhard Ø. Palsson at the University of California, San Diego, in PNAS
2003
Parts that snap together
A short MIT report set out cut sites arranged so that joining any two standard DNA parts leaves a join that can be used again, which made a growing catalogue of interchangeable parts possible.
Credit
Thomas Knight, MIT
2009
One unit for promoter strength
Readings of the same promoter varied by half across seven combinations of strain, growth conditions, plasmid copy number and marker. Dividing each reading by that of one reference promoter cut the variation from 49% and 39% to 17% and gave the field a shared unit, the Relative Promoter Unit. The same paper reports that fewer than 50 of over 500 promoters then in the Registry of Standard Biological Parts had been characterised at all.
Lessona part catalogue is worth little until the numbers in it are comparable.
Credit
Jason Kelly, Drew Endy and colleagues, in the Journal of Biological Engineering
2011
One file format for designs
Version 1.0.0 of the Synthetic Biology Open Language, a file format for exchanging genetic designs between programs, was published on 30 September 2011; it reached version 3.1.0 on 26 October 2022, which is still the current release.
Credit
the SBOL community
2016
Verilog compiled into DNA
Cello takes a circuit written in Verilog, a language used to describe electronic chips, and returns a DNA sequence. The authors designed 60 circuits for E. coli, 880,000 bases in all, built each as the software specified with no tuning, and found 45 worked in every output state, with 92% of all output states as predicted.
Lessona quarter of the compiled circuits still failed, and the authors trace reliability to keeping each gate insulated from its neighbours rather than to the compiler.
Credit
Alec A. Nielsen, Christopher A. Voigt and colleagues at MIT, in Science
2017
A model that computes traits
iML1515 covers 1,516 genes, 2,712 reactions and 1,877 metabolites of one E. coli strain and is free to download from the BiGG Models database, which now lists 108 such models.
Credit
Jonathan M. Monk, Adam M. Feist and colleagues, in Nature Biotechnology
2021
An audit of the parts
A study of the standard-format version of the iGEM parts collection counted 38,464 parts, of which 1,769, or 4.6%, had no sequence recorded at all. It named five problems: little reuse, missing annotation of sub-sequences, incomplete or inaccurate free-text descriptions, duplicate parts, no removal of bad parts, and no continuous curation.
Lessonthe same small core of parts is reused year after year while the rest of the catalogue grows unchecked, so curation has to be a standing job.
Credit
Jeanet Mante, Jacob Beal, Chris J. Myers and colleagues, in ACS Synthetic Biology
Watch for
Circuits designed by a model
A method called CLASSIC reads every circuit in a large library in full with long-read sequencing, tags each with a short barcode, measures how brightly each performs in human kidney cells, and trains a machine-learning model on the result to predict designs that were never built. The team says this is the first use of AI to design genetic circuits and that learned models predicted circuit behaviour more accurately than the physics-based models used until now.
Achieved whena group outside Rice designs circuits from the published model or dataset and reports measured performance that matches the predictions.
Claimed by
Caleb Bashor's laboratory at Rice University, with Kshitij Rai and Ronan O'Connell as first authors, in Nature, January 2026
Open
01
Parts with no measurement
A public table that takes the most reused parts in the standard-format iGEM collection and links each one to every published measurement of it, with the value, the units, the growth conditions and the source, or marks it "no measurement found". Where a paper allows it, convert the value into Relative Promoter Units so rows can be compared.
The briefCredit · unclaimed02
Models of one organism, compared
A dated, rerunnable comparison of every published genome-scale metabolic model of the same organism: which reactions one model carries and another does not, where the gene assignments differ, where the same reaction is written two ways, and one agreement number for each pair of models. Publish the table together with the script that regenerates it.
The briefCredit · unclaimed
Open 01 · the brief
Parts with no measurement
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table that takes the most reused parts in the standard-format iGEM collection and links each one to every published measurement of it, with the value, the units, the growth conditions and the source, or marks it "no measurement found". Where a paper allows it, convert the value into Relative Promoter Units so rows can be compared.
Why it is still open
The 2021 audit of the collection's 38,464 parts found incomplete descriptions, duplicates and no continuous curation, and the 2009 measurement paper found fewer than 50 of over 500 promoters characterised, in units that could not be compared. Neither gap has been filled with a part-by-part record.
Built with what exists
The standard-format iGEM dataset held on SynBioHub; the Europe PMC and Crossref search interfaces; pySBOL3; a current AI model that reads papers into a fixed schema; Zenodo for a dated release.
First step
Download the dataset, rank the parts by how often they appear inside other parts, and publish the top 500 with their current descriptions.
Done looks like
The 500 most reused parts each labelled with at least one cited measurement or with "no measurement found"; a second person re-checks 50 rows; the share of parts with no measurement is published as a single number that later work can argue with.
Open 02 · the brief
Models of one organism, compared
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A dated, rerunnable comparison of every published genome-scale metabolic model of the same organism: which reactions one model carries and another does not, where the gene assignments differ, where the same reaction is written two ways, and one agreement number for each pair of models. Publish the table together with the script that regenerates it.
Why it is still open
The tools are in place and the models are free to download, which makes this a job rather than a research problem. memote's diff command takes two or more models and writes a side-by-side report, and the BiGG Models database holds 108 models, six of them for the same E. coli K-12 MG1655 strain, four for human and four for one Trypanosoma cruzi strain. What has been measured is how far reconstruction software disagrees, not how far the field's curated models do: a 2024 study in npj Systems Biology and Applications built models from the same genomes with three automated tools and found average Jaccard similarity for their reaction sets of 0.23 and 0.24 in two bacterial communities. No published table says how BiGG's own models of one organism differ from each other.
Built with what exists
BiGG's JSON downloads and its public interface; COBRApy; memote and its diff report; Python; GitHub Actions to rerun the comparison when BiGG is updated; Zenodo for a dated release.
First step
Download the six E. coli K-12 MG1655 models from BiGG and publish the lists of reaction identifiers that differ between them.
Done looks like
Every BiGG model for three organisms compared, a table of reactions present in some models and absent from others with counts per model, a published agreement number for each pair, and a script that reruns from one command.