The work done in Alex Norquist’s Chemistry Lab at Haverford College involves both the synthesis of novel compounds with exciting properties as well as addressing inequalities in material chemistry datasets. For instance, some materials can be used for material synthesis but are not due to various human-driven biases and decisions. One example is the material availability bias, referring to the cost of a material, or how readily available the material is to locate and use.
The Norquist Lab at Haverford is divided into several different projects, all pertaining to material chemistry and machine-learning, which can be used to address the inequalities to generate higher quality data. I will be describing the Oxide Project. Yong Suk Choi (‘21), Matthew Danielson (‘21), Joshua Engler (‘22), George Jiang (‘22), Davion Williams (‘23), Teddy Carlin (‘24), and myself are currently working on this project.
While our lab primarily focuses on unbiased exploratory chemistry, a result of exploration is the discovery of new compounds, particularly small, observable single crystals. What are crystals? Crystals and crystalline solids are materials with highly ordered structures. On a microscopic level, they form a repeating structure of units called a lattice. The crystals and crystalline products synthesized in our lab are particularly interesting to study due to their wide structural diversity, as many different elements from the periodic table can be implemented into these structures. By experimenting with new reaction conditions and infrequently used reagents, we may be able to actually generate novel compounds with new bonding, symmetries and shapes.
By experimenting with new reaction conditions and infrequently used reagents, we may be able to actually generate novel compounds with new bonding, symmetries and shapes.
The “Oxide Project” focuses on the unbiased synthesis of “Organically Templated Metal Oxides.” Huh? Let’s break it down. A metal oxide is a compound consisting of two elements: a metal element and oxygen. Metal oxides are unique crystalline solids that form polymeric structures with strong bonding. Organic compounds are chemicals whose structures are largely made of hydrogens and carbons, potentially with some heteroatoms — that is, non-hydrogen, non-carbon atoms — such as nitrogen, oxygen, and sulfur. The introduction of heteroatoms in the structure often determines specific chemical behaviors and properties of the compound.
One common type of organic compounds are amines, which contain nitrogens. In our project, we use different forms of amines as a key reagent in the synthesis of novel compounds. Amines aid in the formation of the crystalline template by situating themselves in special locations in the repeating lattice structure. All together, these compounds are known as organically templated metal oxides.
As much as the Oxide Project group desires to generate crystals, we are also working with combinations of materials, or systems, that are largely unreported in published literature. In fact, recent research conducted by the Norquist Lab suggests the existence of large gaps in the choices of which reagents, more specifically, amines, chemists often use for crystal-forming reactions. In other words, many amines exist that can be used in such reactions, but few actually are. This is a significant finding that needs to be addressed if continuing with exploratory chemistry.
…the Norquist Lab has previously successfully developed a machine-learning program that provides insight into promising reagents that are not found in published literature.
In response to this discovery, our lab generates randomized reactions with unpopular reagents to determine if these materials are just as likely to produce organically templated metal oxides as those frequently reported in published literature. The basic idea is that computer-generated reactions — “randomized reactions” — are free from human bias, which allows us to explore systems that are avoided by most chemists for no legitimate reason, and thus gives us better insight into which reagents may provide successful results. Yet, given that human researchers always have to decide whether to adopt the results of computers or not based on their own judgments, the exploration process can be tricky sometimes.
Using the data from our randomized reactions, the Norquist Lab has previously successfully developed a machine-learning program that provides insight into promising reagents that are not found in published literature. This program incorporates data from combinations of reagents and reaction conditions that resulted in both successes and failures. The machine-learning approach has the capability to save time and resources.
This article was edited by Joe Ding and Anagha Aneesh.