Predict material properties before you synthesize
Matforgelab models how a candidate formulation will behave and ranks it with a confidence interval, so your lab runs the experiments most likely to hit the target.
Materials R&D is still running on intuition and trial.
The average new formulation takes 18 to 36 months from hypothesis to validated candidate. Most R&D cycles spend roughly 70% of that time eliminating compositions that had little chance of working. The information to rank those candidates differently existed in published literature and DFT databases. It just was not connected to the workflow. Matforgelab closes that gap before the lab bench opens.
From composition space to ranked candidates, in hours.
Property prediction engine
Input: composition range and processing parameters. Output: ranked candidate list with per-prediction confidence intervals. The GNN is pre-trained on DFT-computed literature and fine-tuned on your experimental records.
Explore the platformActive formulation search
Bayesian optimization iterates over composition space, proposing the next most-informative synthesis based on predicted performance and current uncertainty. No grid search, no gut-feel.
Characterization ingestion
Upload XRD, DSC, or EIS output files directly. The model reconciles predicted vs measured values and updates its posteriors, so each characterization batch sharpens the next candidate list.
Three material domains, one prediction workflow.
Cathode and electrolyte formulation screening for specific capacity, cycle-life stability, and ionic conductivity. EIS data ingested directly.
Tg, tensile elongation, and solvent resistance prediction across monomer ratio and crosslinker space. Pareto frontier for multi-property trade-offs.
Composition-to-property mapping for fatigue life, yield strength, and corrosion resistance. Narrows coupon screening in the first synthesis round.
Three steps from hypothesis to ranked candidates.
Define your target properties
Set the performance specification: tensile strength range, ionic conductivity floor, glass-transition temperature window. Matforgelab maps your target to the composition-property space and identifies where high-probability candidates are likely to cluster.
Generate and rank candidates
The GNN scores candidate formulations against your target specification. Each candidate carries a predicted property value and a confidence interval. Candidates where the model is confident and the predicted value hits your target go to the top of the synthesis queue.
Close the experimental loop
Upload your XRD, DSC, or EIS characterization results. The model updates its posteriors and the next candidate batch narrows toward your target. Campaign history shows how quickly convergence is happening so you can calibrate synthesis spend against remaining uncertainty.
From R&D scientists who have run early-access campaigns.
We used to spend the first six weeks of any new formulation project just deciding which compositions to synthesize. Matforgelab cuts that to a day.
The uncertainty output is what makes it trustworthy. It tells us where the model is confident and where we need to run the experiment ourselves.