The Sky's the limit for novel proteins

SkyEngine's introduction appeared on Nature's special issue in Biopharma Dealmakers. Please use this link to access the details.

Full Text below:

Cell- and gene-therapy successes are mounting, with a variety of new technologies reaching the market or progressing rapidly through clinical development. But further progress for both therapeutics and delivery technologies will require navigating novel protein sequences based on a broader range of amino acids. To date, pharmaceutical companies remain largely unimpressed with flashy artificial intelligence (AI) tools created for the task and are more interested in results.

During just a few short years, Quantum Sky, based in Cambridge, MA, has developed precisely such a reputation for results. The company was founded by a Boston-based team with complementary AI, synthetic biology, drug discovery and commercial gene-therapy experience. Early accomplishments from its platform, SkyEngine, have inspired Quantum Sky to leverage a combination of protein large-language modeling and molecular-dynamics graph neural-network modeling to generate novel protein sequences for a platform of pipelines for nucleic acid-editing therapies and their requisite delivery vehicles.

Early waves

Quantum Sky was founded in 2022, but SkyEngine had already made its mark near the start of the COVID-19 pandemic. Chengwei Luo, co-founder and CEO of Quantum Sky, deployed the technology through a partnership with Sherlock Biosciences, Feng Zhang’s clustered regularly interspaced short palindromic repeats (CRISPR) company for infectious-disease diagnostics. The SkyEngine prototype designed novel CRISPR-associated protein 12 (Cas12) and Cas13 proteins for real-time disease detection, critical for the development of a COVID-19 diagnostic that secured US Food and Drug Administration (FDA) emergency use authorization in 2021.

“Sherlock continues to use that technology for its diagnostics in development,” said Luo. “It was the first clear demonstration of our ability to discover useful novel protein sequences, but just one of the many actionable applications for SkyEngine. With just 20 amino acids, nature has built intricate protein biomachines.”

Uniquely, Quantum Sky can go even further. “SkyEngine goes beyond the traditional amino acids into all sorts of naturally and artificially modified ones, bringing generative bioengineering a step forward,” said company scientific advisor Alán Aspuru-Guzik, a professor at the University of Toronto, the Canadian Institute For Advanced Research (CIFAR) AI chair at the Vector Institute of Artificial Intelligence, and director of the Acceleration Consortium.

Without a training set of proteins based on the expanded range of amino acids, other AI technologies can’t match SkyEngine, which can find sequences based on how atoms interact within or across proteins and small molecules (Fig. 1). “Quantum Sky has been able to create a system for the ultimate level of protein design, incorporating several hierarchies of detailed information that go beyond traditional approaches,” said Aspuru-Guzik.

Fig. 1 | Advantages of SkyEngine versus conventional large language models (LLMs). a, Rather than providing a static structure that is sampled often and at random, SkyEngine enables conformation tracking of a protein structure with respect to specific conditions, such as ligand binding. b, Computing beyond nature, SkyEngine operates on principles of quantum mechanics and requires only modest training sets; its reach encompasses hundreds of non-canonical amino acids and other molecular types. c, SkyEngine is poised to design a diverse array of biomachines with applications spanning gene and cell therapy. PLP, phage-like particle; VLP, virus-like particle.

The company has developed wet-lab validation capabilities that parallel its AI-discovery engine, allowing it to apply the technology to the therapeutics space. The combination has enabled Quantum Sky to build a pipeline of unique gene editors and delivery systems based on this expanded amino-acid alphabet. Its initial focus is developing therapies for genetic disease, using base editing to correct single-point mutations and RNA writing to correct multiple mutations at once.

“Our strategy is to pursue therapies for diseases with a clear mechanism of action and mature animal models, which will let us address unmet needs for patients while serving as a proof of concept for our capabilities,” said Luo.

That’s just the beginning. Quantum Sky’s RNA-editing capabilities can go beyond traditional genetic diseases to serve patients with more common diseases that may have many genetic contributors, such as cancer or cystic fibrosis. It can also be deployed for gain-of-function mutations, with therapeutic potential in the metabolic disease and immuno-oncology spaces.

Quantum industry entanglement

Quantum Sky operates a hub-and-spoke model, with SkyEngine at its center and each related tool in its own subsidiary, offering potential pharma partners several ways to utilize the company’s capabilities. Some of these subsidiaries, such as Jorna Therapeutics, Inc., have already raised funding to support their activities while Quantum Sky builds toward its own larger funding round.

Validation of the platform has meant continued partnership interest, with an impressive track record that includes diagnostics companies like Sherlock, nonprofits like the Bill & Melinda Gates Foundation, and global pharmaceutical companies like Roche. Many are attracted to the RNA-editor toolset, as well as delivery systems for nucleic acids and proteins that go beyond naturally derived viral vectors.

“Our platform is enabling uniquely cutting-edge technology, like our editors, which couldn’t be developed otherwise,” said Luo. “The differences become evident when our competitors are showing potential partners static pictures of protein structures, while we show them models that are more like interactive movies.”