Opportunities / ENERGY, DEPARTMENT OF

TECHNOLOGY LICENSING OPPORTUNITY: AmineBind ML

Description (excerpt)

A descriptor?based software and model for amine-based carbon capture discovery Organizations that design sorbents for removing CO2 from air gain a fast, chemistry?aware way to rank candidates and focus resources on the most promising structures. AmineBind ML, a trained surrogate model, packaged with user?friendly software, predicts CO2 binding energies for amine active sites from simple molecular inputs. Teams can screen vast chemical spaces in minutes, align material choices with target regeneration temperatures and reduce trial?and?error in lab campaigns. Overview Developed by Los Alamos National Laboratory, the software ingests a chemical structure as a SMILES string, identifies amine binding sites, then uses a descriptor?based machine learning surrogate model trained on roughly 20,000 electronic?structure calculations to predict CO2 binding energetics. Inference runs far faster than density functional theory, which enables high?throughput exploration of millions of candidate chemistries for direct air capture. Predictions at the atomic scale can be combined with mesoscale modeling to feed broader materials pipelines. Technology Description AmineBind ML includes a Python?based toolkit that parses molecular inputs in SMILES format, computes chemically meaningful descriptors for amine sites, and applies a trained model to estimate CO2 binding energies. Training data come from binding energetics computed for ~20,000 molecules, anchoring predictions to first?principles energetics and supporting generalization across diverse amine chemistries. Model inference achieves orders?of?magnitude speed?ups versus DFT, which enables rapid ranking and down?selection prior to expensive simulations or synthesis. This bundle supports screening of millions of structures for direct air capture, delivering candidate materials that balance strong CO2 uptake with manageable regeneration temperatures to minimize operational costs and mitigate sorbent degradation. The atomic?level predictions can integrate with mesoscale treatments, creating a robust modeling pipeline that links molecular binding energetics to process?level performance. Advantages Rapid screening of large chemical spaces from simple SMILES inputs Orders?of?magnitude faster predictions than DFT for CO2 binding energetics Better targeting of materials that balance capture strength and regeneration needs Integration with mesoscale models to support end?to?end materials workflows Software package designed for resear …

View the official notice on SAM.gov →