false false

In the news

IPA’s Subsidiary BioStrand Completes Integration of 20 Million Proprietary Structural HYFT™ and Accelerates in Silico Drug Discovery Developments

Structural HYFT Technology is used as a navigational layer to parse over the protein predictive platforms AlphaFold-2 and Evolutionary Scale Modeling (ESM)-2 Fold

HYFT Technology explores formal and explicit biologically relevant knowledge and connects information about sequence, structure, and function

HYFT Technology is continuously enriched, e.g., by the structural models released by ESM-2 Fold and AlphaFold-2

HYFTs are now linked together by more than 25 billion relationships within the Knowledge Graph (KG)


VICTORIA, BRITISH COLUMBIA (CANADA), December 8, 2022 – IPA (IMMUNOPRECISE ANTIBODIES LTD.) (the “Company” or “IPA”) (NASDAQ: IPA), an advanced biotherapeutic research and technology company, today announced that its subsidiary, BioStrand BV, has completed the full integration of 20 million Structural HYFTs with Deepmind’s AlphaFold and META’s ESM-2 Fold platforms.


Protein Wars:  AlphaFold-2 vs ESM-2 Fold

Recent efforts to predict the three-dimensional structure of a protein using AI are progressing as a solution to the protein folding challenge. A protein’s three-dimensional structure can be studied with the same level of accuracy as experimental methods, thanks to AlphaFold, a protein structure analysis AI created by DeepMind, an Alphabet business focused on artificial intelligence. In July 2021, Alphafold became publicly available and is credited with revolutionizing biology.

A searchable database containing the three-dimensional structures of more than 200 million proteins predicted by Alphafold was released in July 2022.

The ESM Metagenomic Atlas, a database that predicted the structure of over 617 million metagenomic proteins, was then presented in November 2022 by Meta AI, a Facebook research team focused on generating knowledge for the AI community. The study of genomes that have been retrieved directly from environmental samples is known as metagenomics.

ESM is a ground-breaking approach developed by Meta AI researchers to predict protein structure. This model is one of the closest alternatives to DeepMind’s AlphaFold-2, which reportedly resolved the 50-year-old grand challenge of in silico protein folding. Meta AI has introduced several models throughout the years, and the public can now see its most recent efforts. ESM-2 Fold inference is quicker at enabling the investigation of structural spaces of metagenomic proteins, even though AlphaFold-2 and RoseTTAFold have higher accuracy.

Numerous protein structure prediction models exist in addition to ESM-2 Fold and AlphaFold-2, such as RoseTTAFold, OmegaFold, IntFOLD, RaptorX, and others.

The ESM Fold model from Meta’s protein structure prediction AI translates the atoms and molecules that make up the protein into language and predicts the three-dimensional structure from the learning data. The group developed the ESM-2 with 15 billion parameters by extending this model. ESM-2, the largest protein language model to date, employs 2,000 GPUs, and is able to predict the three-dimensional structure of more than 600 million proteins in the ESM metagenomic atlas in just two weeks.

Meta AI’s research team claims that while ESM-2’s structure prediction speed is up to 60 times faster than AlphaFold’s, its prediction accuracy is inferior to that of AlphaFold. This suggests that structure prediction can be scaled to considerably larger databases, according to Meta.

The HYFT Technology

HYFTs are Universal FingerprintTM patterns mined throughout the whole biosphere. When linked together, they form a Knowledge Graph that constitutes over 660 million HYFTs and more than 25 billion relations. The core characteristic is that these HYFTs can connect sequence to structure and function, but also link sequence to all types of textual information such as scientific papers and medical records. Recently, the Company also added more than 20 million structural HYFTs (S_HYFTs) to this Graph, and continuously adds metadata and relations. This strengthened the HYFT-based platforms with the double compounded effect of harnessing the structural prediction capabilities of AlphaFold-2 and ESM-2 as navigational layers for the HYFTs to parse over, while integrating the associated knowledge and speeding up the discovery processes.

Continuously enriching and updating the HYFT graph with the latest novelties is a core characteristic.  This means that the number of relationships within the graph is exponentially growing. It provides the HYFT-based LENSai platforms with constantly updated and integrated knowledge relevant within a biological context.

What makes this graph unique is the wealth of explicit information on the whole biosphere that is represented.  Since graphs hinge on relationships, they allow one to easily and efficiently determine and visualize the connectedness of different entities in this biosphere.  Furthermore, it equips the LENSai platform with an extremely powerful starting point for further application of various AI and machine learning techniques in antibody discovery and precision medicine. The HYFT graph provides a unique intermediate layer between sequences and predictions, allowing new knowledge and insights to be extracted without having the downside of operating in a black box.

From Knowledge Graph to Undiscovered Knowledge 

Why does it matter to have a huge knowledge graph covering the whole biosphere? The richness of a graph determines the depth and level of detail of new information and knowledge that can be extracted.  Having access to a wealth of 25 billion relationships empowers fine-grained levels of exploration that were not possible before. It opens a whole new world of undiscovered knowledge. In addition to the magnitude of relations, the biological relevance of the information is key. Here, the HYFT knowledge graph’s unique ability to connect sequence, structure, function, and literature is unprecedented.

In Silico and Wet Lab Integrated

BioStrand’s AI-driven approach and IPA’s best-in-class laboratory capabilities are leveraging our target-agnostic antibody discovery platform to create the best possible drugs in a variety of therapeutic domains. Supporting the wet lab with the most up-to-date and complete information and predictions using a wide range of in silico models accelerates wet-lab experiments, and generally makes them more efficient and insightful.


ImmunoPrecise Antibodies Ltd.

ImmunoPrecise Antibodies Ltd. has several subsidiaries in North America and Europe including entities such as Talem Therapeutics LLC, Biostrand BV, ImmunoPrecise Antibodies (Canada) Ltd. and ImmunoPrecise Antibodies (Europe) B.V. (collectively, the “IPA Family”). The IPA Family is a biotherapeutic research and technology group that leverages systems biology, multi-omics modelling and complex artificial intelligence systems to support its proprietary technologies in bioplatform-based antibody discovery. Services include highly specialized, full-continuum therapeutic biologics discovery, development, and out-licensing to support its business partners in their quest to discover and develop novel biologics against the most challenging targets. For further information, visit www.ipatherapeutics.com.


Investor contact: investors@ipatherapeutics.com



Forward Looking Information


This news release contains forward-looking statements within the meaning of applicable United States securities laws and Canadian securities laws. Forward-looking statements are often identified by the use of words such as “potential”, “plans”, “expects” or “does not expect”, “is expected”, “estimates”, “intends”, “anticipates” or “does not anticipate”, or “believes”, or variations of such words and phrases or state that certain actions, events or results “may”, “could”, “would”, “might” or “will” be taken, occur or be achieved. Forward-looking information contained in this news release includes, but is not limited to, statements regarding the effect of the integration of 20 million structural HYFTs to the Knowledge Graph on structural prediction capabilities and speed of the discovery processes, as well as statements relating to the expected outcome of integrating in silico models and wet-lab experiments. In respect of the forward-looking information contained herein, IPA has provided such statements and information in reliance on certain assumptions that management believed to be reasonable at the time.


Forward-looking information involves known and unknown risks, uncertainties and other factors which may cause the actual results, performance or achievements stated herein to be materially different from any future results, performance or achievements expressed or implied by the forward-looking information. Actual results could differ materially from those currently anticipated due to a number of factors and risks, including, without limitation, the risk that the integration of the above mentioned HYFTs and the integration of in silico models and wet-lab experiments may not have the expected results, as well as those risks discussed in the Company’s Annual Information Form dated July 28, 2022 (which may be viewed on the Company’s profile at www.sedar.com), and the Company’s Form 40-F, dated July 29, 2022 (which may be viewed on the Company’s profile at www.sec.gov). Should one or more of these risks or uncertainties materialize, or should assumptions underlying the forward-looking statements prove incorrect, actual results, performance, or achievements may vary materially from those expressed or implied by the forward-looking statements contained in this news release. Accordingly, readers should not place undue reliance on forward-looking information contained in this news release. The forward-looking statements contained in this news release are made as of the date of this release and, accordingly, are subject to change after such date. The Company does not assume any obligation to update or revise any forward-looking statements, whether written or oral, that may be made from time to time by us or on our behalf, except as required by applicable law.



SOURCE ImmunoPrecise Antibodies