How AutoRxn works

From reaction idea to optimised conditions, in four clear steps.

AutoRxn sits between your chemistry and your automation. Define the space, upload your results, and let Bayesian optimisation decide which experiments are worth running next.

Designed for chemists, not data scientistsWorks with plates, vials, and flow
Reaction optimisation loop
  1. 1Define factors, ranges, and objectives.
  2. 2Run reactions and upload results.
  3. 3AutoRxn learns the response surface.
  4. 4New batch of experiments proposed.

A loop built around real experiments

1
Define

Set up your reaction space

Choose your substrates, catalysts, solvents, and ranges for key parameters like temperature, time, and equivalents. AutoRxn builds the design space for you.

  • Continuous and categorical factors
  • Hard limits, safe ranges, and exclusions
  • Multi-objective or single-objective targets
2
Measure

Upload or connect your data

Import reaction tables from ELNs, spreadsheets, or automation platforms. Keep yields, selectivity, and analytical data aligned to each experiment.

  • CSV / XLSX imports (structured columns)
  • Outlier flagging and data validation
  • Support for multi-response outputs
3
Model

Model the response surface

AutoRxn uses Bayesian optimisation to learn how your factors influence performance, with explicit uncertainty estimates for each region.

  • Gaussian process models under the hood
  • Uncertainty-aware predictions
  • Live diagnostics and model fit
4
Optimise

Generate smarter experiments

Select acquisition strategies (exploration vs exploitation) and AutoRxn proposes the next batch of experiments for your reactor or platform.

  • Batch proposals matched to plate or reactor formats
  • Safety constraints and practical limits respected
  • Track improvement over baseline conditions

Under the hood: Bayesian optimisation, exposed just enough.

AutoRxn uses probabilistic models to balance exploration and exploitation. You get sensible defaults out of the box, with the option to tune strategies when you want more control.

Models

Gaussian processes with kernels tailored to chemical design spaces, including categorical factors.

Acquisition

Expected improvement and exploration-heavy strategies for early campaigns, with batch selection support.

Constraints

Apply lab and safety limits directly in the UI so suggested experiments stay physically realistic.

Ready to run your next optimisation loop in AutoRxn?

Start with a pilot project, or connect your existing reaction tables to see how the modeller behaves on real data.