The paper demonstrates how library-based molecular data storage is able to address some shortcomings of DNA-based data storage. Most notably, Gumus uses waste chemicals to store data, demonstrating how library-based molecular data storage can both reduce costs and environmental impact compared to DNA-based storage.

The team behind AtomICs developed a scheme to turn acid-base systems into logic gates needed to perform computational tasks. This scheme was then used to construct a neural network which successfully classified images of handwritten numbers with over 98% accuracy. With continued research into chemical computation, it may become possible to use molecules to both to store data and to run computational tasks.

In addition to benefits such as resource savings and increased spatial information density, molecular data storage benefits from the ubiquity of chemicals, which enables use cases such as chemical steganography. In particular, the team behind AtomICs successfully encoded information on a dollar bill by manipulating the particular arrangement of naturally occurring chemicals, effectively hiding data in plain sight.

Researchers with AtomICs developed a theoretical framework to introduce a means of non-linear computation into chemical computation. These non-linear computations are crucial for neural networks, such as image classification programs. Using the theoretical framework, researchers successfully constructed a neural network to classify animal silhouettes with >80% accuracy.

The AtomICs team lays out a number of information storage schemes and analyzes each of the schemes. The paper provides the formulas necessary to calculate the theoretical information capacity of a small molecule storage system given a set of constraints and investigates the benefits and tradeoffs associated with different storage schemes.

Building off of earlier work, the AtomICs team demonstrated a new means of library creation using Ugi products and ran a successful proof of concept. This demonstrates the viability of non-biologic small molecules as a medium of information storage as well as offering the potential for faster and cheaper library synthesis.

The paper unpacks recent developments regarding molecular computation and discusses some ongoing challenges around realizing a molecular computational scheme.

The team behind AtomICs successfully ran a proof of concept, encoding multiple images in molecular mixtures and then reading them back out, mimicking the information storage functions of a biologic metabolome. Although completed on a relatively small scale, the test proved that the concept was viable and revealed avenues for improvement as AtomICs continues to refine the technology.

The team behind AtomICs develops the theoretical framework necessary to develop a synthetic chemical perceptron, a foundational unit of computation in neural networks. This paves the way for further work both to expand upon the theoretical foundations established here and to develop a working chemical perceptron.