Our approach

Better starting points. Smarter decisions. Less wasted lab time.

Most drug discovery starts in vast, random chemical space. Ours starts with molecules that microbes have spent a very long time refining, then uses ApaisBio's AI to decide what's worth making and testing.

01 / The thesis

A weak starting point is the most expensive mistake in discovery.

If you've run a discovery program, you've felt it: months of synthesis and screening before you learn a series was never going to work. Bigger libraries don't fix that on their own. Better starting points, and better ways to rank them, might.

That's the thesis behind ApaisBio's metabolic programs, which pair our technology partner's discovery platform with ApaisBio's own AI-coordinated discovery.

02 / The discovery engine

The discovery engine: how our molecules are found and designed

01 Access

Access

Our technology partner's Industrial-Scale Microbiome Mining (ISMM) harvests and concentrates natural microbial communities at kilogram-to-ton scale. That scale can surface rare, low-abundance small molecules that standard culturing or small-sample methods often miss.

02 Read

Read AI

Each harvest is profiled with deep analytical chemistry (LC-MS/MS), genomic, habitat, and biological-activity data. Together these form training sets for ApaisBio's AI models, which are designed to connect chemistry to biology.

03 Prioritize

Prioritize AI

ApaisBio's AI-coordinated discovery ranks candidates by biology, novelty, and feasibility, using spectral matching, clustering of unknown molecules, structure prediction from analytical data, and activity and target hypotheses, and building on our technology partner's analytics.

04 Design

Design AI

AI in-silico screening and AI-assisted medicinal chemistry explore derivatives and analogs designed for the properties a gut-restricted oral drug needs, such as stability in the stomach and staying in the gut. Early AI-driven liability and ADME/tox triage is designed to flag risks before anything is synthesized.

05 Validate

Validate AI

Only top candidates move into the lab: wet-lab confirmation, safety screens, and animal studies, each set up as a go/no-go gate. Results feed back into our AI models to guide the next round of optimization. If a program doesn't clear a gate, we stop or redesign rather than spend more.

03 / Two ways to start

Typical discovery vs. our approach

  1. 01Starts in random synthetic chemical spaceStarts with biology-shaped chemistry from microbial communities
  2. 02Screens more molecules to find a hitUses AI to rank fewer, more meaningful candidates
  3. 03Learns about liabilities late, after synthesisDesigned to flag liabilities in silico, before synthesis
  4. 04Measures success by throughputMeasures success by meaningful candidates per unit of effort

04 / What makes this different

A targeting system, not a single target.

Most drug programs pick one target, push potency there as hard as possible, and deal with side-effect liabilities later. We map the lead target together with the cooperating pathways that help and the liabilities that hurt, then score each candidate against an ideal window instead of a single number.

A single-target compound can look strong on its main target and still fall outside that window, because it ignores the rest of the system. Our goal is a profile that stays high where it helps and low where it harms.

  • Ideal window
  • Single-target approach
  • Coordinated targeting profile
Illustrative schematic only; not experimental or predicted data.

Engine 1

Microbial discovery

We look for cryptic biosynthetic pathways: gene clusters that microbes carry but rarely switch on in the lab. A pathway moves forward only when published, peer-reviewed biology suggests that kind of hidden chemistry is relevant to the pharmacology we want. That keeps us from chasing attractive structures with no biological footing.

Chemistry from real biological relationships. These pathways can sit inside a single microbial genome, or across a holobiont: a host together with its microbes. There, the chemistry is part of a real biological relationship, shaped over time by the host and its microbes working together. We treat that relationship as a clue to what a molecule is for.

Engine 2

Synthetic design

Chemists and in-silico tools then tune those starting points for potency, oral exposure, stability, and balance across the targeting map, so a candidate is designed to sit inside the window rather than merely score high on one target. ApaisBio's AI ranks, docks, and optimizes candidates against that map and within the geofence; it is a method for deciding what to make and test, not a result. Each candidate is also designed to stay in the gut (see pharmacological geofencing).

05 / Eroom's Law

Why it matters

For decades, the cost of bringing each new drug to market has kept rising, a trend often called Eroom's Law (Moore's Law spelled backward). Our founder's research argues that one cause is starting chemistry spread thin across random chemical space. Our goal is to change that math: start from chemistry that biology has already shaped, and find far more meaningful candidates for every dollar and week spent.

This describes our design aim, not results. ApaisBio has an exclusive relationship in metabolic disease and obesity across our technology partner Biosortia, Inc.'s BIOS1 library.

Read the science

White papers and preprints are not peer-reviewed publications.

Books by our founder: the Breaking Life Sciences series

These books reflect the author's perspective. They are not independent scientific validation and are not medical advice.

AI that improves specificity, not just throughput.

ApaisBio's AI-coordinated discovery spans AI in-silico screening and AI-assisted medicinal chemistry and optimization. More molecules isn't the goal; sharper decisions, made earlier, are. As our metabolic programs grow, we plan to build the next stage of our discovery stack on Google Cloud.

Planned

Vertex AI

We plan to extensively use Vertex AI to train, tune, and deploy our molecule-ranking and property-prediction models, and to run larger in-silico screens, such as binding and liability predictions, on demand.

Planned

Gemini

We plan to extensively use Gemini models to help our scientists synthesize literature, patents, and assay results faster, and to generate and stress-test hypotheses before they reach the bench.

Planned

Google Cloud data platform

We plan to extensively use Google Cloud to bring spectral, genomic, and assay data into one governed, secure environment, so our models and our scientists work from the same source of truth.

These describe planned infrastructure, not current deployments. Google, Google Cloud, Vertex AI, and Gemini are trademarks of Google LLC; their mention does not imply endorsement or partnership. Multiple large language models and drug discovery platforms augment our efforts.

Protected science, shared the right way.

Patent applications pending.

We share the opportunity publicly and keep the map private. Molecule structures, data, and search strategy are available to qualified partners under confidentiality and other agreements.

Want to go deeper?

See where each metabolic program stands on the pipeline, or start a conversation.

Partnering and investment inquiries

Contact ApaisBio

Let's start a conversation.

Partnering, licensing, investment, or research collaboration. Only your email is required.

Please send information about