The Problem: Behavior Data Is Missing

Science can sequence a gene or model a protein's 3D structure. But it cannot cheaply predict how a molecule will behave when it hits stomach acid, a mixing vat, or a patient's bloodstream. This gap kills drugs in late-stage trials and forces food companies into endless recipe tweaks.

Apoha, a London-based startup spun out of 15 years of interfacial physics research, claims to have built the missing measurement. On June 3, it emerged from stealth with $36M in funding, announced at SXSW London's Frontier Technologies Stage. The round is led by Singular, with participation from Tim Draper's Draper Associates and seed investors Redalpine, Seedcamp, Wilbe, and Nucleus, plus Innovate UK grants.

What Apoha Built: VIBE and Liquid State Intelligence

Apoha calls its data layer "Liquid State Intelligence" — a new category alongside sequence and structure. Its first product is VIBE, an empirical readout of how a sample behaves under controlled stress.

Here's how it works: You take a quantity of material small enough to sit on a pinhead (about 8 micrograms). Suspend it in liquid. Apply a sequence of perturbations — mechanical or acoustic stress. Record the wave patterns the molecule throws off in response.

Those patterns resolve into more than 1,000 measured descriptors of behavior in a single reading. Conventional assays capture one property at a time. VIBE does it in minutes.

Technical Details: The Physics Behind It

The science traces to 2008, when founder and CEO Shamit Shrivastava began working on a problem left open by the Nobel-winning Hodgkin-Huxley model of nerve signaling: the physics of the boundary where matter meets liquid. In 2014, he published evidence for two-dimensional solitary sound waves at a lipid interface, work later named among Scientific American's discoveries that could change everything. The company now holds more than 60 patents across hardware, software, data, and AI models.

The key insight: conventional assays measure static properties (e.g., viscosity at rest). VIBE measures dynamic response — how the molecule's surface behaves under stress, which correlates with real-world performance.

Benchmark Results: >90% Precision on Antibody Candidates

Apoha's strongest evidence is a preprint from joint research with Boehringer Ingelheim, a multi-year commercial partner. VIBE identified high-risk antibody candidates with greater than 90% precision from as little as 8 micrograms of material.

A second version of the benchmarking work reports the platform outperforming 12 industry-standard developability tests across 236 clinical antibodies. Importantly, VIBE surfaces information the conventional measures miss rather than duplicating them — meaning it catches failures that standard assays would miss.

Commercial Traction and Use Cases

VIBE is already in commercial use. Customers span pharma, biotech, and food:

  • Boehringer Ingelheim: multi-year commercial partner for antibody developability.
  • Ethris: German biotech using VIBE to predict how lipid nanoparticles carrying mRNA behave in animals.
  • THIS: plant-based food company using VIBE for a protein replacement bound for supermarket shelves.
  • Somru BioSciences and several Fortune 500 companies across pharma, food, and materials.

Why This Matters for Developers

Physical-world AI — systems that act on matter — is a growing field. Models can now see and read, but none can feel how a drug dissolves or how a flavor holds. That data has never been collected at scale. As Shrivastava said: "It cannot be scraped from the internet, synthesized, or retrofitted from existing assays. It has to be measured."

For developers building AI models in drug discovery, materials science, or food tech, Apoha offers a new data class. Instead of training on static properties, you can train on behavioral fingerprints. This could dramatically reduce the data needed to predict real-world outcomes.

Next Steps

Apoha will use the $36M to scale VIBE deployment and build out the Liquid State Intelligence data layer. The company is hiring across hardware, software, and data science roles. Developers interested in working with behavioral data at scale should watch for API access — the company has not announced a public API yet, but the data layer suggests one is coming.

If you're building ML models for physical-world applications, start thinking about how behavioral data could complement sequence and structure. The gap Apoha is filling is real, and the benchmarks suggest it works.