Possible use of Research Papers to aid in simulations

Materials Project
1

Hi everyone,

I am completely new to both the Materials Project and materials science, so please assume I may be misunderstanding some fundamentals.

I have been wondering whether there is any research or ongoing work that combines large-scale materials literature, such as research papers, with machine learning models for property prediction.

My thought was that there are millions of published materials science papers containing information about compositions, processing conditions, and measured properties. If that information could be extracted and structured, perhaps it could be used to train models that learn relationships between composition, processing, and material properties, and thereby help predict the properties of new materials.

For example, could information from the literature be combined with datasets such as those in the Materials Project to help estimate the properties of new alloys, composites, or materials that have not yet been experimentally characterized?

I realize that this may be an oversimplification, since material properties depend on much more than composition alone, including processing history, microstructure, phase behavior, and other factors. I am mainly trying to understand whether this general idea is realistic and whether similar efforts already exist within the materials informatics community.

I also wonder whether such an approach could complement the Materials Project by helping guide computational exploration toward promising new materials or compositions.

Any pointers to existing projects, papers, or reasons why this would or would not work would be greatly appreciated.

2

I have been wondering whether there is any research or ongoing work that combines large-scale materials literature, such as research papers, with machine learning models for property prediction.

My thought was that there are millions of published materials science papers containing information about compositions, processing conditions, and measured properties. If that information could be extracted and structured, perhaps it could be used to train models that learn relationships between composition, processing, and material properties, and thereby help predict the properties of new materials.

Yes, there is a substantial body of research focused on analyzing the scientific literature to predict material properties, synthesis methods, and material characteristics. Even before the development of modern LLMs, researchers were applying traditional NLP techniques to extract and leverage knowledge from the literature. In parallel, machine learning models have also been widely used for materials discovery, simulation, and design.

For example, could information from the literature be combined with datasets such as those in the Materials Project to help estimate the properties of new alloys, composites, or materials that have not yet been experimentally characterized?

The Materials Project aims to provide high-quality, well-curated materials data that is grounded in physical principles and generated through highly accurate simulations performed and validated by domain experts.

See more here!

3

Thanks for the quick and insightful response :+1: