by 
melon frost
Sam Levin’s grandfather was a fifth-generation potato farmer in western Massachusetts struggling with falling potato prices at Midwestern industrial farms. After trying to grow a range of other crops, a set of melon seeds smuggled from Puerto Rico seemed to do the trick, thriving in the region’s sandy loam soil. According to family legend, these melons were just ripe for harvest when a frost struck and killed the harvest.
Levin is now the CEO and co-founder of Melonfrost, a Brooklyn-based evolution startup that combines its proprietary software and hardware to drive evolution in an automated closed loop. The technology aims to provide a novel way to develop and produce new microbes on a large scale for everything from food and energy to therapeutics and synthetic materials – all part of not designing or engineering the future, but rather, growing it. “Our aim is to no longer be completely at the mercy of catastrophes like frost, but to be able to produce frost-hard melons for any purpose, figuratively speaking,” he explains.
Machine learning scientists meeting in office
Various tools for growing microbes with specific traits for desired uses have historically been limited by the ability to scale – creating a bottleneck in the transition from an engineered strain to a commercialized one – with many methods of doing so relying on relatively expensive and somewhat brute-force estimates -and-check approaches, typically based on mutations in a genetic sequence. Instead, Melonfrost’s recent $7 million seed round, co-led by Refactor Capital and Alexandria Venture Investments, supports a thesis that, as Levin puts it, “evolution has been and will continue to be the finest designer of organisms for a long time .
Central to this focus on phenotype selection are Melonfrost’s Evolution Reactor hardware and Maia, its proprietary software platform. Maia is a suite of machine learning algorithms that learn how organisms evolve — in terms of different selection pressures and environmental conditions relative to measured phenotypes — and iteratively return a set of instructions in the form of further selection pressures to further evolve a desired set of traits , whether yield or frost resistance. These input and output data connect Maia to the evolutionary reactor, the apparatus for individually controlling, measuring, and applying these encoded selective pressures to cultivate thousands of independent microbial populations on parallel evolutionary paths.
Large-scale control of evolution is made possible by a series of hardware innovations encapsulated in a series of modular units in the evolution reactor, each containing approximately 250 individual microbial populations. The two platforms, virtual and mechanical, are woven together by cloud software that closes the loop of the automated evolution control platform – data fed into the software from the hardware, instructions returned to the hardware through modeling software updates – which iterate until the desired phenotype goal is reached or the loop is turned off. Currently, the entire system just about fits into the Brooklyn lab at Melonfrost, but Levin articulates the vision for this hardware-software interface as a “biological data center” in the form of an evolutionary reactor warehouse.
Scientists working in the laboratory
This seed round is the next step towards the full form of this evolutionary control system – funding the next phase of building the Evolution Reactor hardware and bringing Melonfrost closer to its first customer in the food sector edible fats. “Feeding the world without destroying it in the process is a particular area of synthetic biology where there are many bottlenecks to go from initial construction to production,” Levin points out. This focus on building a healthier world through food isn’t new to Levin and his co-founder, Head of Engineering & Design and childhood friend Loren Amdahl-Culleton. In high school, the duo started a farm for their cafeteria to increase student investment in the learning community and work toward sustainability. Despite spending undergraduate and high school years separated by an ocean and an entire country, the two from Oxford to Stanford stayed in touch as they studied evolutionary dynamics and reinforcement learning, respectively, and began noticing the potential to fill gaps in evolutionary models with machines fill learning tools that share similarities in their underlying mathematics. With two other childhood friends, Melonfrost was born—driven by the positive impact of synthesis in areas ranging from cutting-edge machine learning and hardware engineering to synthetic biology and precision custom software tools.
“Each of these endeavors would require a lot of expertise, failure, and innovation, so doing all of these at the same time is a bit unusual,” Levin admits, “but these challenges are so great that you have to innovate on multiple fronts at once and integrate a lot different kinds of scientists and engineers to really make the future grow. It’s not just about bringing new molecules or chemicals to market; Rather, we need to fundamentally change the way the world’s resources are created and moved.” For Melonfrost, the goal is not to eventually build large factories and ship them in shipping containers. Instead, the vision is to quickly, cheaply, and robustly create and optimize new strains — extending to production in general, to translate from learning the language of evolution to reliable large-scale biological outcomes — no matter the metaphorical freeze may come.
Many thanks to Aishani Aatresh for additional research and reporting on this article. I am the founder of SynBioBeta and some of the companies I write about are sponsors of SynBioBeta Conference and weekly summary.
