Our Technolgy

SuntheticsML is a machine learning platform for chemistry optimization that recommends experiments based on existing lab data. Users upload a CSV or Excel file exported from a notebook or ELN, and the model begins learning from as few as five data points without requiring data cleaning scripts or code. Each column in the dataset is labeled as continuous, discrete, categorical, or chemical, and the model handles each type accordingly to ensure recommendations correspond to experiments that can actually be run. Multiple objectives such as purity and yield can be optimized simultaneously, with user-defined weighting and targets. Constraints including arithmetic, Boolean, and N-choose-K rules define which variable combinations are permitted, filtering out infeasible conditions before they reach the bench. SuntheticsML automatically tests multiple model types and selects the best fit for each output, though users can manually specify a model such as a Gaussian process. Each predicted optimum includes a 95% confidence interval. Suggested experiments are presented as a ranked shortlist that scientists can edit, skip, or supplement with their own runs. When results return from the lab, new data rows are added and a new model version is trained on the full accumulated dataset. SHAP values identify which input variables most influence each outcome. A Pareto front visualization shows trade-offs between competing objectives and displays all predictions in context to support candidate selection.