Target-specific AI/ML workflows for oral-small-molecule discovery

ArrePath moves from target selection to validated, deployed target-specific AI/ML workflows in about four weeks, then explores novel chemistry at machine scale.

Define the indication first, then design the molecule

Potency, oral bioavailability, ADME/PK, and cytotoxicity are designed in parallel against an indication-specific candidate profile.

ArrePath indication-first design process: define the target product profile, build and validate target-specific AI/ML workflows, then design and test molecules while optimizing potency, oral bioavailability, ADME/PK, and cytotoxicity in parallel.

Three ways the platform delivers

~4 weeks
Rapid program start
Target selection to validated, deployed target-specific AI/ML workflows
1M+
Machine-scale exploration
Molecules evaluated per design cycle
Multi-parameter
Developability from day one
Potency, oral bioavailability, ADME/PK, and cytotoxicity optimized in parallel

Ariadne: evidence-powered scientific copilot

Ariadne connects proprietary experimental data, public drug-discovery data, literature and patent evidence, chemistry design, and predictive models to turn scientific questions into evidence-backed answers with citations.

Ariadne evidence map: research questions flow through proprietary experimental data, predictive models and analytics, chemistry design and sourcing, literature and patent evidence, and public drug-discovery data to produce evidence-based answers with citations.

Design across the whole candidate profile

Production models span potency, clearance, solubility, plasma-protein binding, cytotoxicity, oral bioavailability, and related PK properties. These properties are considered together from the start rather than retrofitted after potency optimization.

ArrePath production-model breadth across potency and developability, including antibacterial potency, clearance, solubility, plasma-protein binding, cytotoxicity, oral bioavailability, and related pharmacokinetic properties.

Heterogeneous data is an asset, not noise

Most scientists and ML practitioners discard substantial amounts of bioactivity data because absolute values do not reproduce across assays. ArrePath learns the reproducible signal across those assays instead.

  • Learn rank orders, not absolute values — they reproduce across assays
  • Use scaffold-agnostic molecular representations — generalize beyond a chemical series
  • Result: models that zero-shot generalize to new chemistry

Our models generalize where public models fail

Predict oral bioavailability of our lead-series compounds — our model had no training data on this chemical series.

  • ArrePath model: AUC-ROC ≈ 0.72 (zero-shot on the lead series)
  • Public Random Forest: ≈ 0.56
  • Public GNN: ≈ 0.51 (essentially random)

Cross-scaffold, cross-assay rank-order training is what enables generalization.

Bar chart of zero-shot oral bioavailability classification AUC-ROC on Project 1 (n=70), bootstrap mean ± 1 SD — ArrePath model approximately 0.72, public Random Forest approximately 0.56, public GNN approximately 0.51.

One platform, multiple programs

Reusable models, consistent data treatment, and target-specific workflows support programs across the updated pipeline.

Strategy & Pipeline →