Amii and New Harvest use machine learning to speed up cultivated meat research
Amii and the non-profit New Harvest have published two papers that apply machine learning to cellular agriculture, the production of food such as meat by growing animal cells directly. One builds an open dataset on growth factor heat tolerance, the other uses Bayesian optimization to find better culture media, in some cases in as few as half the trials, according to Amii.
- Amii and New Harvest published two papers applying machine learning to cellular agriculture.
- One paper builds a dataset on the heat tolerance of growth factors used in cultured meat.
- The other uses Bayesian optimization to find culture media formulations, sometimes in as few as half the trials.
- Data and models are intended to be open for the wider field.
- The papers belong to the second phase of the Amii and New Harvest collaboration.
Two papers, one collaboration
The research builds on a long-term collaboration between Amii and New Harvest, a non-profit dedicated to advancing cellular agriculture. Amii says there has been limited collaboration and cross pollination between the cellular agriculture and machine learning communities so far.
Both papers are meant to be open resources for the field, in line with New Harvest's open-source mandate and Amii's philosophy of "AI for good and for all". Bre Duffy, Director of Responsible Research & Innovation at New Harvest, said the projects address "the two main barriers to realizing ML's potential in cellular agriculture: a lack of public data and researchers with expertise in both domains."
Amii describes cell-based agriculture as an emerging technology with the potential to address challenges in the global food industry, including sustainability, food-borne diseases and ethical concerns around animal husbandry.
A dataset on growth factor heat tolerance
The first paper, Growth factor thermostability dataset derived from experimental and computational approaches to accelerate cultured meat development, was published in June. Co-authors include Kiana Aghakasiri, a University of Alberta student working with Amii, Dave Staszak and Breanna Duffy. Food System Innovations, a non-profit investing in solutions for more sustainable food protein, funded the work.
The team combined laboratory testing, computer modelling and LLM-assisted research reviews to compile data on the heat tolerance of growth factors used in cultured meat. Growth factors act as signals that tell cells to grow and differentiate into structures such as muscle and fat. They can come from plant- and animal-based sources, and different ones break down at different cooking temperatures.
Knowing how growth factors react to temperature is critical for food safety and food regulation, according to the article. By gathering data from different sources into one dataset, the researchers say they offer a tool for the wider scientific community.
Finding better culture media faster
The second paper, A data-driven evaluation of optimization techniques in cell culture media, was published last month. It was co-authored by Staszak, Ali Parsaee, formerly a Machine Learning Resident at Amii, and Miranda Stahn, formerly New Harvest's Program Manager, Western Canada. Schmidt Sciences, a philanthropic initiative founded by Eric and Wendy Schmidt, supported it.
The work targets culture media, the nutrients in which cells are grown. Amii says finding the best mix of nutrients and growth factors is extremely difficult, with billions of possible combinations, and that it is one of the most costly parts of the process.
Using a form of Bayesian optimization, the team built a model that predicts which culture media will be more effective for a specific production. According to Amii, the model found optimal formulations much faster than traditional statistical methods, sometimes in as few as half the trials. Amii says this kind of optimization could dramatically lower production costs and speed up project timelines.
What it means for the field
Staszak called the two projects "very practical, applied problems" that will help solve two bottlenecks holding the technology back from larger adoption: cost efficiency and safety.
Both papers belong to the second phase of the collaboration. The first phase reviewed current research on machine learning in the field, while the second brought in experts to apply machine learning to major challenges in cellular agriculture. Because the data and models are open, other researchers can use them beyond this collaboration.
