MEMo AI: Predicting Mixture Properties via Molecular Interaction

스마트경제 South Korea

Overview

Professor Jun-young Jung’s research team at Kookmin University developed MEMo, an AI model that predicts the properties of chemical mixtures by learning intermolecular interactions. Validated against 860,000 data points from NIST, the model outperforms existing methods in predicting density, sound speed, vapor pressure, refractive index, and surface tension.

In Depth

Key Findings

The MEMo (Mixture Embedding via Mole Fraction) AI model demonstrates superior predictive accuracy for chemical mixture properties compared to existing representative methods. In a validation study using over 860,000 experimental data points from the U.S. National Institute of Standards and Technology (NIST), MEMo showed significantly higher accuracy in four out of five key properties: density, sound speed, vapor pressure, refractive index, and surface tension. The performance gap was particularly pronounced in mixtures with strong intermolecular interactions, such as ionic liquids and salt solutions, as well as mixtures containing molecules not previously encountered during the AI’s training phase.

Technical Details

Unlike conventional AI models that analyze individual molecular characteristics and calculate a simple weighted average based on mixing ratios, MEMo integrates “which molecule is mixed and in what proportion” from the initial stage. The architecture consists of a front-end that identifies the inherent characteristics of each molecule and a back-end where different molecules exchange information to learn real-time interactions. This allows the AI to recognize the same molecule differently depending on its mixing partner and ratio, capturing non-linear changes in properties that traditional averaging methods miss.

Background and Context

Chemical products, ranging from industrial solvents and battery electrolytes to cosmetics, are typically complex mixtures. Because properties change non-linearly based on mixing ratios, verifying every possible combination through physical experimentation is practically impossible. The research, conducted by Professor Jun-young Jung along with integrated MS/PhD student Ji-hwan Kim and undergraduate researcher Chan-jung Kim, was published in the SCI-level journal Chemical Engineering Journal (Impact Factor 13.2, JCR Q1) under the title “Encoding composition-dependent intermolecular interactions into learned molecular representations.”

Significance and Outlook

MEMo offers high versatility, as a single model can process everything from pure substances to complex mixtures of two or three components. To support a transparent academic ecosystem, the research team has released the evaluation data and model code. This technology is expected to reduce the time and cost of research and development in fields such as battery electrolyte design, eco-friendly solvent development, and carbon dioxide capture materials.

Get our weekly technology intelligence — free

Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.

Subscribe Free — Weekly Tech Intelligence

By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.

• Your email and selected fields are used only to deliver the newsletter.
• We never share your information with third parties.
• You can unsubscribe anytime via the link in each email.

See our Privacy Policy for details.

Agree & Continue ▶

Takes about a minute · Unsubscribe anytime