Using a pre-trained AI model from NVIDIA, the startup Evozyne has created two proteins with significant potential in healthcare and clean energy.

A joint paper describes the process and the biological building blocks it created. One aims to cure a congenital disease, while the other is designed to consume carbon dioxide to reduce the effects of global warming.

“It’s really encouraging that even in this first round, the AI model has created synthetic proteins that are as good as naturally occurring ones. This tells us that it has properly mastered the rules of nature’s design”,

says Andrew Ferguson, co-founder of Evozyne.

A transformative model of artificial intelligence
Evozyne uses NVIDIA’s implementation of ProtT5, a transformational model that is part of NVIDIA BioNeMo, a software framework and service for creating AI models for healthcare.

“BioNeMo really gave us everything we needed to support training the model and then running tasks with the model very inexpensively – we could generate millions of sequences in just a few seconds.”

says Ferguson, a molecular engineer working at the intersection of chemistry and machine learning.

The model is the basis for Evovyne’s process, called ProT-VAE. It’s a workflow that combines BioNeMo with a variational autoencoder that acts as a filter.

The model studies the ways nature
Like a student reading a book, NVIDIA’s transformational model reads amino acid sequences across millions of proteins. Using the same techniques that neural networks use to understand text, it learns how nature assembles these powerful building blocks of biology.

The model then predicts how to assemble new proteins suitable for the functions Evozyne wants to address.

Deep neural networks have emerged as an important type of AI model, capable of achieving state-of-the-art performance across natural language processing, computer vision, and other domains. The strength of these models lies in their hierarchical structure, which generates a large amount of highly parallelizable work well-suited for multicore hardware like GPUs. Frameworks for general-purpose GPU computing such as CUDA and OpenCL have made the development of high-performance programs easier in recent years. Yet GPUs remain especially challenging to optimize, in part because their architectures rapidly evolve.

Domain-specific languages and compilers have emerged to address the problem, but these systems tend to be less flexible and slower than the best handwritten compute kernels available in libraries like cuBLAS, cuDNN, or TensorRT. Reasoning about all these factors can be challenging even for seasoned programmers. The purpose of Triton, then, is to automate these optimizations, so that developers can focus on the high-level logic of their code.

A key challenge however is that batteries degrade quickly with operating conditions. It is currently difficult to estimate battery health without interrupting the operation of the battery.

In work recently published by Nature Machine Intelligence, researchers developed a new method to estimate battery health irrespective of operating conditions and battery design or chemistry, by feeding artificial intelligence (AI) algorithms with the raw battery voltage and current operational data.

Darius Roman, PhD student that designed the AI framework commented:

“To date, the progress of data-driven models for battery degradation relies on the development of algorithms that carry out inference faster. Whilst researchers often spend a considerable amount of time on model or algorithm development, very few people take the time to understand the engineering context in which the algorithms are applied. By contrast, our work is built from the ground up. We first understand battery degradation through collaborations with the CALCE group at the University of Maryland, where batteries’ in-house degradation testing was carried out. We then concentrate on the data, where we engineer features that capture battery degradation, we select the most important features and only then we deploy the AI techniques to estimate battery health.”

Besides, the researchers found that current data-driven models for battery health estimation do not consider model confidence. However, this is often critical for decision-makers to understand how the AI model came to a certain conclusion and whether the model can be trusted. In their work, the proposed AI model is capable of quantifying uncertainty in its predictions to better support operating decisions.

The developed framework scales up with new chemistries, including the new upcoming solid-state batteries, battery designs and operating conditions and has the potential to unlock new strategies of how batteries can and should be used.

According to Valentin Robu, from the Smart Systems Group:

“Batteries are increasingly critical to a variety of applications, from robotics to renewable energy integration. A key challenge in these domains is having accurate, high-confidence estimates of battery state of health. Consider, for example, a robotic asset operating in a remote environment such as deep subsea monitoring, were assuring the health of the battery deployed on the robot is mission-critical.”

Valentin also added that the process is similar for energy applications, having an accurate estimate of the remaining useful lifetime of the battery is often critical to a project’s economic viability.