AI-powered software for chemical process development that uses Bayesian Optimization and Machine Learning to accelerate reaction optimization for pharmaceutical and biotech companies.
Our Mission:
We aim to accelerate and automate chemical process development by equipping wet-lab chemists with the power of data-driven optimization and robotic execution of experiments.
The Problem:
The discovery of novel pharmaceuticals is one of our most important weapons in fighting disease. However, the drug development pipeline is often held up for many months during the design of chemical processes to manufacture these drugs at scale, delaying FDA trials and lengthening the time until drug launch. Designing chemical processes involves the identification of suitable parameters such as catalyst/temperature/solvent. Currently process development is often done via tedious trial-and-error experimentation (slow) or exhaustive screening (expensive and wasteful).
Our Approach:
In our research, we have developed algorithms for chemical process optimization, which leverage transfer learning and Bayesian optimization. We validated the algorithms in the wet lab, showing an up to 95% reduction in required experiments and cost compared to exhaustive screening. We have made our approaches accessible to chemists through our user-friendly no-code software platform and to automated laboratory equipment with our API.
An optional in-app AI assistant powered by Large Language Models on Amazon Bedrock that summarizes results, suggests next steps, and helps propose parameter spaces.
What customers say
“Validated in industrial case studies, our robust optimization algorithms consistently deliver measurable performance gains.”
CustomerReactWise customer
“Built on proprietary datasets and descriptor databases, ReactWise uniquely integrates chemical intelligence into every prediction.”
CustomerReactWise customer
“Connection with laboratory instruments for real-time, closed-loop optimization.”
CustomerReactWise customer
Optimization Platform
On Demand Pharmaceuticals used ReactWise's Bayesian optimization platform and visual analytics to guide an API synthesis campaign. ReactWise reports 50% fewer experiments than a conventional campaign would require and a…
ReactWise reports using iterative multi-objective Bayesian optimization with Pfizer to optimize a hydrogenation temperature profile. The vendor reports 97.5% yield, 1% impurity and identification of a temperature-profile…
ReactWise Launches Explainability and Robustness Tools for Process Scale-Up
ReactWise has updated its platform to include deeper model explainability features. The new capabilities allow users to identify parameters driving yield and impurity, visualize robust operating regions, and simulate the effects of process deviations. These tools are designed to facilitate tech transfer, regulatory submissions, and scale-up decision-making by providing a mechanical understanding of optimal process conditions rather than relying solely on experimental optimization results.
New feature
ReactWise Launches MemoryBO Transfer Learning for API Process Development
ReactWise has introduced MemoryBO, a transfer learning algorithm that integrates with Electronic Laboratory Notebooks (ELN) and data lakes. This feature allows users to leverage historical campaign data to inform new process development experiments. By structuring existing experimental data, the platform enables teams to carry forward knowledge from previous work on related intermediates, reducing the need to restart investigations for similar chemical processes.
New feature
ReactWise Updates Kinetic Modeling Workflow with New Visualization and Analysis Features
ReactWise has released an update to its kinetic modeling workflow that enables users to group time series data by experimental runs and visualize kinetic curves side by side. This improvement facilitates rapid comparison of species behavior across different temperatures and concentrations. By streamlining the visualization and grouping process, the update allows users to move directly into kinetic model identification using the platform's Kinetics360 tool, reducing the time required for data manual preparation.
