[Paper Review] Machine Guided Discovery of Novel Carbon Capture Solvents

James L. McDonagh, Benjamin H. Wunsch+6 more|arXiv (Cornell University)|Mar 24, 2023

Machine Learning in Materials Science4 citations

TL;DR

This paper presents an end-to-end machine learning-driven discovery pipeline for novel carbon capture solvents, combining a high-throughput lab assay with a molecular fingerprinting model to predict CO2 capture capacity. The method achieved 60% accuracy on external test sets for multiple parameters and identified previously untested amines with high experimental validation, including a new open-source tool and a curated dataset for future research.

ABSTRACT

The increasing importance of carbon capture technologies for deployment in remediating CO2 emissions, and thus the necessity to improve capture materials to allow scalability and efficiency, faces the challenge of materials development, which can require substantial costs and time. Machine learning offers a promising method for reducing the time and resource burdens of materials development through efficient correlation of structure-property relationships to allow down-selection and focusing on promising candidates. Towards demonstrating this, we have developed an end-to-end "discovery cycle" to select new aqueous amines compatible with the commercially viable acid gas scrubbing carbon capture. We combine a simple, rapid laboratory assay for CO2 absorption with a machine learning based molecular fingerprinting model approach. The prediction process shows 60% accuracy against experiment for both material parameters and 80% for a single parameter on an external test set. The discovery cycle determined several promising amines that were verified experimentally, and which had not been applied to carbon capture previously. In the process we have compiled a large, single-source data set for carbon capture amines and produced an open source machine learning tool for the identification of amine molecule candidates (https://github.com/IBM/Carbon-capture-fingerprint-generation).

Motivation & Objective

• To address the high cost and time burden of traditional materials development for carbon capture solvents.
• To develop an integrated machine learning and experimental pipeline for rapid screening of novel amine candidates.
• To create a reliable, reproducible assay for measuring CO2 absorption capacity in aqueous amines at scale.
• To generate a large, single-source dataset of amine capture performance for training and benchmarking ML models.
• To release an open-source machine learning tool for molecular fingerprinting to accelerate future solvent discovery.

Proposed method

• A high-throughput lab assay measures CO2 absorption capacity in 130 aqueous amine solutions at 30% w/w concentration using infrared detection at 4.3 µm.
• The assay quantifies CO2 capture capacity (α) via integration of time-resolved gas concentration signals, with calibration using MEA as a reference (α = 0.50 mol CO2/mol amine).
• A machine learning model uses molecular fingerprints to predict CO2 capture performance based on molecular structure.
• The model is trained on a curated dataset of 76 amines from prior literature and validated on an external test set of 23 benchmark compounds.
• The pipeline integrates experimental screening with ML prediction in a closed-loop discovery cycle to prioritize high-potential candidates.
• The open-source tool (https://github.com/IBM/Carbon-capture-fingerprint-generation) enables molecular fingerprinting and candidate prediction for new amines.

Experimental results

Research questions

• RQ1 Can a machine learning model accurately predict CO2 capture capacity in aqueous amines using molecular fingerprints?
• RQ2 Can a high-throughput experimental assay reliably measure CO2 absorption capacity across diverse amine structures?
• RQ3 Does the integration of experimental data and ML prediction enable the discovery of novel, high-performing amines not previously considered for carbon capture?
• RQ4 How well does the model generalize to external test sets, particularly for amines with non-traditional reactivity?
• RQ5 Can the discovery pipeline reduce the time and cost of solvent development compared to conventional screening?

Key findings

• The machine learning model achieved 60% accuracy on an external test set for predicting multiple CO2 capture parameters.
• For a single parameter, the model reached 80% accuracy on the same external test set.
• The experimental assay showed moderate correlation (R² = 0.67) with reference data from Puxty et al. for 23 benchmark amines.
• The pipeline successfully identified and experimentally verified several novel amines not previously applied to carbon capture, including L-leucinol and tert-butylaminoethanol.
• Capture capacity trends matched expected reactivity: α ≈ 0.5 for primary/secondary amines and α ≈ 1.0 for tertiary amines, with outliers like piperidine (α = 0.93) and 2-amino-2-methyl-1-propanol (α = 0.90) observed.
• The study compiled a large, single-source dataset of 130 aqueous amine solutions with measured CO2 absorption performance, now publicly available with the open-source tool.