Technology
aiProtein®: A robust integration of Directed evolution with Artificial intelligence
Proteins, which are key molecules for biological function in all organisms are translated from genetic information encoded in DNA. They are generated as linear polypeptides of 20 naturally occurring amino acids, then form three-dimensional structures for biological function. The number of the sequences, for example, of proteins composed of 300 amino acids is 2x10390, which is extremely larger than the number of stars in space, 1023. To explore and find a novel protein sequence in this vast space, we need an unprecedented and phenomenal way.
RevolKa's strategy
RevolKa developed a proprietary technology to explore the vast sequence space, called aiProtein®. This is achieved by integrating directed protein evolution with artificial intelligence (AI). RevolKa is committed to quality biological data with less noise and errors, which accurately trains our machine-learning engines. aiProtein® does not require big data, but relatively smaller numbers of data for accurate prediction. This saves time and labor significantly, and opens opportunities for complex biological data, such as cell cytotoxicity and receptor internalization. aiProtein® has been highly validated with different types of proteins, such as antibodies and enzymes.
Conception of aiProtein®
The core concept of aiProtein® was created in an academic collaboration between four Japanese renowned scientists in protein engineering and artificial intelligence (AI). Prof. Umetsu at Tohoku university is a pioneer of next-generation directed protein evolution. He is also a chief scientific officer at RevolKa. Prof. Tsuda at The University of Tokyo, Dr. Kameda and Dr. Saito at the National Institute of Advanced Industrial Science and Technology are top scientists in theoretical information technology and AI. They teamed up for the integration between protein engineering and AI.
ACS Synth. Biol. (2018) 7, 2014-2022
ACS Catal. (2021) 11, 14615-14624
Collaboration Project for machine-learning-guided molecular evolution
Tohoku University
Professor M. Umetsu
Molecular evolution
The University of Tokyo
Professor K. Tsuda
Machine-Learning
T. Kameda, Ph.DMachine-Learning
Professor Y. Saito
Machine-Learning
*The National Institute of Advanced Industrial Science and Technology (AIST), one of the largest public research organizations in Japan.
External advisor
The University of Tokyo
Professor K. Tsuda
Scientific Advisor
Tohoku University
Assoc.Prof. H. Nishi
Scientific Advisor
Tohoku University
Assoc.Prof. H. Nakazawa
Technical Advisor
Artificial intelligence
AI-assisted technology, aiProtein® enables us to overview fitness landscapes of sequence spaces and guides us to travel over the landscape to find promising variants with extraordinary properties. Classical directed protein evolution repeats search for promising variants in the vicinity of a starting protein. aiProtein® has lower risks to stall at a semi-optimal peak and is capable to go beyond deep valleys.
Essential technologies
aiProtein® has two key steps. First, the vast sequence space of a particular protein is going to be narrowed down to an appropriate size to explore. Strategies for this space extraction are based on RevolKa’s knowledge and experience. Second, data sets of variants randomly selected from the extracted space are generated to collect rough information about the landscape. Our machine learning engine is trained with the data and models a landscape to propose a list of candidates to test. Harmonization of these two steps is a very fundamental technique in aiProtein®.
Multi-dimensional protein engineering
Multi-dimensional engineering of protein properties is one of the advantages of aiProtein®. Each property has an inherent landscape in the sequence space. Simultaneous engineering of multi-properties was a challenge in the classical directed protein evolution. Machine-learning in aiProtein® can import training data on multiple properties to model a landscape that meets all of them. In fact, RevolKa successfully improved three properties simultaneously.
Tecnhnical validation
Transformation of GFP to YFP
We transformed a green fluorescence protein (GFP) to a yellow fluorescence protein (YFP) by using aiProtein®. ~100 variants data were generated to model the landscape for yellow fluorescence. Our machine-learning engine proposed a ranked list of GFP variants. Some of the variants showed brilliant yellow color, the brightness of which was strongest in the YFP variants in the world.
Creation of Super enzyme
We created a super enzyme that has superior enzymatic activity by using aiProtein®. ~100 variants data regarding enzymatic activity were generated to model a landscape for the performance factor. Our machine-learning engine proposed a ranked list of the enzyme variants. A variant showed 5 times higher activities than the wild-type enzyme. We repeated aiProtein® with the superior enzyme. As a result, some of the variants identified showed improved activities, but suggested saturation of evolution. This indicates that aiProtein® can achieve high-level evolution without iteration.
To be disclosed
Opportunities
Fine-tuned properties
aiProtein®’s multi-dimensional engineering capability enables fine engineering of multiple protein properties simultaneously. This technology provides us with a remarkable directed protein evolution to improve and maintain properties, and even reduce liability. We believe aiProtein® is a powerful tool for many industries, such as pharmaceuticals and chemicals.
Acceleration of R&D and Creation of values
aiProtein® technology is a solution for transformative innovation of biologics: Extraordinary values and competitiveness, developability, and manufacturability. For example, directed design toward subcutaneous administration and room temperature shelf life can enable unprecedented molecular creation. Revolka leverages aiProtein® technology to accelerate the discovery and development of biologics and contributes to scientific research and the health and well-being of people in the world.
All biologics modalities
aiProtein® is a powerful engineering technology for all of the molecules composed of amino acids, such as enzyme, IgG, diabody, bispecific antibody, peptide, etc. RevolKa has successfully engineered a variety of modalities, suggesting significant potential of aiProtein® for molecular creation. Our technology keeps growing through building our know-how and skills. We commit to further technological evolution and challenging new modalities, such as gene therapy.
Partnership
Leveraging our proprietary technology, aiProtein®, RevolKa accommodates all industrial demands on proteins, such as therapeutics, foods, and chemicals. We are always open to research collaboration with partners who can share our passion to challenge the frontier of protein engineering. Please contact us for more details.