Sculpting Evolution
Advancing Biotechnology Safely
We are committed biotechnologists who:
Catalyze beneficial advances by applying robotics and machine learning to evolve new molecular tools and techniques
Apply molecules, models, and cryptography to defend against pandemics and prevent the catastrophic misuse of biotechnology
Work with the guidance of interested communities to safely and humanely edit wild populations and ecosystems
To learn more, dive into our research or publications, meet the group, read our philosophy, or donate to support our work.
Contact: sculpting-admin[at]media.mit.edu
Our goal is to thoughtfully ascend the tree of knowledge. We must accelerate our harvest of beneficial technological fruits to sustain, protect, and improve civilization, while refraining from exploring branches harboring advances so powerful and accessible that they pose catastrophic risks. In other words, we must learn to sculpt the evolution of biotechnology.
Recent Lab News
12 December 2024
On the feasibility and risks of mirror life. Policy Forum, Technical Report, and plan for inclusive future discussions.
May we be proven wrong.
22 October 2024
Joanna, Zac, and the rest of the Mice Against Ticks team figured out how to engineer white-footed mice... which necessitated unraveling most of their reproductive biology.
Why Sculpt Evolution?
Evolution gave rise to every living thing and all of human culture, but evolved systems are very different from those designed by humans. They're harder to predict and to design, and exhibit a frustrating tendency to evolve away from engineered behaviors. At the same time, harnessing and directing evolution can generate useful organisms and technologies that we could never have rationally designed.
Our laboratory seeks to understand why systems evolve in the ways that they do, to develop tools capable of precisely intervening in the evolution of ecosystems, and to cultivate wisdom sufficient to know whether, when, and how to proceed.
Lab News
27 August 2024
How can we best prevent pandemics? Despite the controversy over sampling viruses in animals and studying potential pandemic pathogens in the laboratory, no one has generated a quantitative model of the expected benefits of these two strategies - until Geetha's manuscript today, which translates expert opinions from a survey of One Health and vaccine researchers into bounded expected lives saved. Despite wide variation, virus discovery and sequencing appears markedly superior to characterization (0 to 1.46 million lives versus 10,800 to 96,000 lives), largely because experts believe that any given identified pandemic-capable virus probably won't be the one to cause the next pandemic.
15 May 2024
Reliable AI will eventually safeguard humanity, but in the interim, we need to prevent models from providing information that could help create pandemic-class biological weapons, or design worse ones. We've previously found in the classroom and in user studies that models provide some useful information, but struggle to provide overall utility. Of course, everyone expects future models to be more capable. Responsible reporting and relevant benchmarks, like those released today, can help.
18 April 2024
We previously wrote about safety and security concerns involving transmissible vaccines. That's why I'm especially grateful to Daniel Streicker and Scott Nuismer for their kind invitation to join their NSF-funded workshop on the subject. After many thoughtful discussions, we agreed on guidelines for the field, now published as a Policy Forum. While I care most about the commitment to avoid increasing transmissibility - which could be generalized to human viruses - the overall recommendations are fantastic. It's a model example of scientists with different perspectives coming together to find a shared path forward.
27 March 2024
Surprisingly, most ML biodesign tools are trained on datasets without accounting for experimental uncertainty, which is like ignoring error bars. To solve this, Vikram Sundar has created FLIGHTED, a Bayesian approach to modeling sources of experimental error and incorporating them into probabilistic fitness landscapes for training. Intriguingly, it changes relative model performance across standard benchmarking tasks.
6 November 2023
We convened researchers from diverse fields to build a research agenda comprehensively examining the safety of far-UVC germicidal light, now published in Photochemistry & Photobiology. Huge thanks to everyone who came together, especially the folks at Longview who helped organized the event, David Sliney for his stellar advice, and Lenni and Max for following up with the group and leading the writing!
26 October 2023
Covid-19 was terrible, but we shouldn't simply prepare for the last war. What if the next pandemic virus is much more transmissible, higher-lethality, or both? What if it spreads just as rapidly, but causes few symptoms until months or years later, like HIV? These are certainly possible -and our investigation and roadmap strongly suggest that we can reliably defend against them if we prepare.
20 October 2023
Rick Wierenga and Stefan Golas published their stellar work on PyLabRobot, an open-source framework permitting liquid-handling robots and other lab equipment to be programmed in Python, in Device. Together with Wilson Ho in Conor Coley's group, who integrated Tecan, they've created a platform and an associated forum to accelerate laboratory automation.
7 August 2023
Congratulations to Emma Chory, who is now starting her own lab at Duke!
The first postdoc to join our group, she's been integral to nearly every project we've pursued over the past few years. Check out her website to see her amazing research plans.
We will miss Emma terribly, as evidenced by the fact that no fewer than five members of our lab will have visited her there by the end of August.
Congratulations as well to Stefan Golas, our amazing former lab technician, who will be joining Emma's group to start his PhD!
14 November 2022
We become scientists because we want to understand the world. One day, that will allow us to predict which particular virus, chimera, or combinatorial set of mutations will cause another pandemic. That could help us target vaccines... but it will also allow anyone with access to the relevant synthetic DNA and common laboratory equipment to cause new pandemics at will. How should we prepare for that future? A roadmap.
30 December 2021
Our phage-and-robotics-assisted near-continuous evolution (PRANCE) platform for systematic directed evolution was published in Nature Methods. The original PACE system is about evolving biomolecules rapidly; PRANCE lets us do it in parallel, evolving up to 96 populations at once while monitoring the activity and changing conditions as needed.
Erika conceived PRANCE and built the early platform, while Emma ran many of the core evolution experiments and fine-tuned it. It wouldn't have been possible without key assistance from Dana, Stefan, and Brian, plus contributions from many other lab members.
8 December 2021
Kevin testified before the House Foreign Affairs subcommittee on Asia, the Pacific, Central Asia, and Nonproliferation's hearing on "Biosecurity for the Future". Key point: A credible pandemic-capable virus is a credible and accessible weapon of mass destruction. Pandemic virus prediction, in assembling a list of such viruses in order to prevent natural pandemics, will unavoidably give thousands the power to simultaneously ignite as many pandemics as would naturally occur in a century. His written testimony is available here.
30 November 2021
We now have a somewhat functional if interim website following the design of Google Sites Classic. We're still working on finding a professional to redesign it for us - if you're enthused by our work, please contact sculpting-admin[at]media.mit.edu.
We're excited to welcome several new faces. Vikram Sundar joins us from the Computational and Systems Biology program, and will be focusing on evolutionary simulations and machine learning models. Summer DeAmelio joined us as a full-time lab technician in June, and has delighted everyone by gifting the lab with twice-weekly nanopore sequencing runs. Christy Dennison, an expert machine learning engineer formerly of OpenAI, has joined us as an M.Eng. student. William Bradshaw is a former collaborator on bidirectional contact tracing and genetic attribution who will be focusing on future biosecurity and the Nucleic Acid Observatory. Zac Hill, our new Director of Transgenesis, will be working closely with Joanna on Mice Against Ticks and Sebastian on Project Rarity, as well as generating new strains to test daisy drive approaches. In February, we will welcome Devanand Bondage, a specialist in mammalian engineering and host-pathogen interactions.
We're deeply grateful to Bill Lombardi for helping keep everything running smoothly through the pandemic. We are now looking to hire a lab Chief Operating Officer, an Executive Assistant, and multiple Research Scientists for the Nucleic Acid Observatory.
14 November 2021
Let's assume there's a good chance that thousands of people will be able to start new pandemics within a decade. Several folks in our lab can generate infectious samples of many viruses from a genome sequence, so this seems sadly plausible. What to do?
Delay proliferation and misuse to buy time
Detect subtle threats reliably and early
Defend by blocking infections outright
Pandemic proliferation a solvable problem. We now have a roadmap detailing what to do. The question is whether we'll actually do it.