AI platform 'can predict feed trial results'
Umami says Virtual Marine Cell means physical trials only need to be used on the highest-potential candidates
Marine biotechnology company Umami Bioworks has announced a “major advancement” in its ALKEMYST artificial intelligence platform, introducing computational feed biology that predicts how fish cells metabolise nutrients, lipids, and algal oils before physical feed trials begin.
According to Umami, the new capability both extends the power of the ALKEMYST platform and marks a meaningful shift in how feed decisions are made in aquaculture.
The company says as supply of fish oil used in feed tightens, algal oil has become a critical omega-3 source, but that its performance remains difficult to predict. Most formulations rely on trial-and-error: long test cycles, high capital and animal use, and limited visibility into why a formulation succeeds or fails, says Umami. Core biological processes such as lipid metabolism, nutrient uptake, immune activation, and stress response have historically been nearly impossible to anticipate in advance.
Feed performance can be computed ahead of trials, shifting feed development from empirical testing toward a predictive, biology-first approach
“Umami’s Virtual Marine Cell closes this gap by modelling fish biology at the cellular level. Powered by ALKEMYST, the platform simulates how different species process proteins, lipids, and algal oils under varying environmental conditions,” the company says in a press release.
“This allows feed performance to be computed ahead of trials, shifting feed development from empirical testing toward a predictive, biology-first approach.”
Umami says ALKEMYST can now simulate how marine cells metabolise algal lipids, revealing optimal inclusion rates, metabolic constraints, and biological trade-offs that were previously hidden. This enables feed producers to reduce waste, manage costs, and target specific fatty acid profiles linked to growth, resilience, and product quality. This capability directly addresses some of aquaculture’s most persistent challenges: uncertainty around algal oil performance across species and life stages, over-formulation of costly lipid ingredients, long and capital-intensive trials with limited biological clarity, and inconsistent outcomes in growth, health, and fillet quality.
Evaluated 'in silico'
With the Virtual Marine Cell, dozens of feed formulations can be evaluated ‘in silico’ within hours, says Umami, allowing physical trials to focus only on the highest-potential candidates.
“By computing feed biology at the cellular level, we understand how fish respond to nutrition before committing time, capital, and animals. That changes how decisions are made,” said Shivansh Singhal, machine learning scientist at Umami Bioworks.
With feed accounting for up to 70% of production costs in species such as salmon, tuna, and yellowtail, the ability to predict nutritional outcomes represents a structural improvement in aquaculture economics, says the company, which adds that it is already working with global aquaculture producers and feed companies to deploy this capability across species, formulations, and production systems.
