A materials discovery engine built around your experimental loop.
Property prediction, Bayesian candidate search, and characterization ingestion in one place. Built for the bench scientist who needs ranked candidates with confidence intervals, not a pipeline to wire together from scratch.
Request AccessFrom data in to ranked candidates out.
Every discovery campaign passes through three layers: data ingestion, prediction and active search, then workflow output. Each layer updates the next in a closed loop.
Six capabilities, one discovery workflow.
Graph neural network property models
Compositions and crystal structures are encoded as graphs: atoms as nodes, bonds as edges. The GNN learns structure-property relationships from DFT-computed literature and experimental records, generalizing across crystal systems without hand-crafted descriptors.
Bayesian active search
Expected Improvement acquisition function proposes the next synthesis candidate from unexplored composition space. Campaign-level convergence is tracked across rounds, so you see how quickly the model is narrowing toward your property target.
Characterization file ingestion
Upload XRD, DSC, EIS, or TGA output files. Measured results update the model's posteriors in batch or streaming mode, so each experimental round makes the next candidate list more accurate.
Per-prediction confidence intervals
Ensemble and Monte Carlo dropout methods output a confidence interval on every property prediction. You see exactly where the model is reliable and where you need the lab to decide.
REST API and Python SDK
Call the prediction engine from your existing ELN, LIMS, or automation scripts. A Python import, not an IT project. Documented API reference included at all plan levels.
Composition-property dashboard
Interactive scatter plots across your composition-property space, ranked candidate shortlists, and full campaign history. Export to CSV or PDF for internal review and lab scheduling.
Fits where your data already lives.
Matforgelab is designed to fit into the data formats and protocols R&D labs already use. If you have data in a CSV, an ELN export, or a Python script, you are two steps from a ranked candidate list.