Platform

AI, Rosetta, and lab-based large-scale screening.

Cyrus’ platform combines laboratory-tested AI and Rosetta computational methods with a full biochemistry lab specializing in quantitative, large-scale mutagenesis scanning and state-of-the-art protein engineering and assaying techniques.

Computational protein design has evolved tremendously over the last 20 years, and especially in the past five. Rosetta was the first design software to realize most major protein design achievements, such as modifications of stabilization, affinity, enzyme specificity, the creation of novel assemblies, and even the experimentally-validated design of entirely new structures with no sequence similarity to natural proteins.

Taking advantage of significant advances in protein AI, we have built a highly efficient, integrated and proprietary software platform employing the most successful design tools, building novel biologics through a focus on re-engineering existing proteins, and rapid feedback between modeling and lab testing at Cyrus in vitro and in vivo.

Our Capabilities

Structure prediction

We use a range of AI and Rosetta tools for protein structure prediction, which have been deployed for difficult to solve structures with dozens of companies including Nimbus and Selecta. Our CSO Yifan Song led key innovations at the IPD in structure prediction at the semiannual Critical Assessment of Structure Prediction (CASP).

Immunogenicity reduction

We have a track record of reducing T- and B-cell (antibody) mediated protein immunogenicity. The Cyrus team pioneered key machine learning methods in T-cell immune epitope modification, and has worked on immunogenicity reduction for gene therapy candidates in DMD, novel reduced immunogenicity CRISPRs with the Broad Institute, and other programs where highly immunogenic proteins are being developed as potent therapeutics. T-cell epitope reduction uses Cyrus’s proprietary iterative algorithm & large-scale MHC display platform.

Stability and serum half-life

We employ a variety of protein fusions and modifications as well as computational approaches to achieve protein stabilization. We computationally modify protease cleavage sites, and we use other methods to achieve superior stability and serum half life. These methods have proven critical in achieving preclinical objectives in our lead IdeS and ACE2v2 COVID programs.

Solubility & reducing aggregation

The solubility of protein therapeutics can impact both the route of administration and required dosing levels, and protein aggregation can influence the cost and efficiency of manufacturing and drug efficacy. Cyrus’s proprietary ML methods have been validated multiple times for the in vitro modification of solubility and aggregation propensity.

Affinity

Efficacy is often a direct function of the ability of the drug to find and bind tightly to its target. Cyrus uses a variety of Rosetta, AI, and deep mutagenesis laboratory methods, iterating between in silico design and in vitro testing, to tune protein/protein binding affinities. These methods have yielded multiple program leads including our optimized ACE2v2 COVID program, which required optimization in manufacturability, resistance to SARS-CoV-2 spike mutations, and spike affinity – all in one molecule.

Specificity

Poor specificity may mean a drug binding both its desired target as well as to related but undesirable targets, limiting efficacy and potentially causing side effects. Cyrus uses a combination of Rosetta, AI, and deep mutagenesis evaluation of both desired and undesirable targets to identify novel sequences with improved specificity.

Cyrus Deep Mutagenesis Technology Provides Superior Data

Cyrus proprietary methods permit the Deep Mutagenesis Scanning (DMS) of every possible amino-acid substitution in a protein, using a low noise and highly physiologically relevant approach.
DMS can be used to tune affinity and specificity in concert with Cyrus design algorithms.
Cyrus’s DMS has shown the ability to identify valuable gain of function mutations not identified by other groups.
We focus DMS on specific protein families as part of our design strategy, yielding a unique and proprietary data set and a material per-target engineering advantage over pure computational approaches.

T he image on the right shows published deep mutational scan binding affinity data of Cyrus’s ACE2.v2 COVID prophylaxis asset with the Receptor Binding Domain of SARS-Cov2 spike protein.

Cyrus + Rosetta AI/ML

Rosetta software developed at the UW IPD combines statistics and atomistic physics. Many Cyrus scientists were core developers of Rosetta.
Over the past 15 years, machine learning has been added to complement Rosetta, and for the past 4 years the Rosetta community has led in protein design and prediction using AI or Deep Learning algorithms.
The Cyrus team are long-term ML developers for protein design including the creation of the original T cell epitope deimmunization algorithms in Rosetta.
Deep Learning/AI algorithms now play a key role in many of our internal and partnered projects, including post-translational modifications, solubility, stability predictions.
In 2022 Cyrus co-founded the Openfold consortium with Amazon, NVIDIA, Genentech, Outpace, and Arzeda. Lucas Nivon, CEO at Cyrus is a member of the 4-person OpenFold Executive Committee, responsible for choosing new research directions in protein AI.

Rosetta & IPD: Key Firsts in Protein Design

David Baker’s IPD and Rosetta have achieved all of the major firsts in protein design, from the very first accurate structure predictions to the latest design methods for de novo protein design.

De novo Binder Mimics, HA, Neurotoxin B

Homology Modeling

First De Novo Protein

Protein/Protein Interface Design

First De Novo Enzyme

De novo Binder Mimics, HA, Neurotoxin B

Homology Modeling

“I like doing things that seem like magic.”

David Baker

Founder of the Institute for Protein Design and co-founder at Cyrus