Axial partners with great founders and inventors. We invest in early-stage life sciences companies such as Appia Bio, Seranova Bio, Delix Therapeutics, Simcha Therapeutics, among others often when they are no more than an idea. We are fanatical about helping the rare inventor who is compelled to build their own enduring business. If you or someone you know has a great idea or company in life sciences, Axial would be excited to get to know you and possibly invest in your vision and company . We are excited to be in business with you - email us at info@axialvc.com
Who leads Deep Genomics?
Deep Genomics was founded in 2014 by Brendan Frey, Andrew Delong, and Hui Yuan Xiong to bring the power of deep learning technology and talent to drug development. The company is centered around its AI Workbench platform to develop new medicines pursuing genetically-informed targets. Through the combination of deep learning with genomics, high throughput screening, and automation, Deep Genomics generates large biological data sets to validate new genetic targets and develop a corresponding therapeutic.
What does Deep Genomics do?
The company discovers and validates new drug targets and develops medicines to inhibit, and sometimes activate, the target. The first component is led by Deep Genomics’ cornered resource of deep learning talent with the second part mainly achieved with antisense oligonucleotides (ASO).
Deep Genomics measures itself by its success rate of converting a research project to a drug lead (they claim over 70% success rate so far) and the time it takes to do this (aiming to get this process to under a year). The company’s platform has the ability to search over 2,000 diseases across more than 100K pathogenic mutations to discover new drug targets. Deep Genomics sources information from datasets for protein-DNA/RNA interactions, RNAi screens, genomics, chromatin accessibility assays, and more. Once the disease-causing mechanism is pinpointed, Deep Genomics screens for lead candidates to pursue their internal list of targets.
The company’s lead program is in Wilson Disease, a rare disease (~1/30K incidence rate) characterized by excessive copper in the liver, brain among other essential organs, and serves as a case study on Deep Genomics’ platform. Its platform discovered a mutation in M645R (found in 1/50 of patients with the disease) that led to a loss-of-function (LoF), through exon skipping, in the copper-binding protein, ATP7B. After this discovery, several thousand compounds were screened to restore ATP7B levels. The company’s model identified ATP7B (M645R) where others have missed it because this variant is still functional but not expressed due to exon skipping. By training their model to look for splicing problem solely from sequence data, the company then conducted a screen against the mutant ATP7B honing in on 12 lead candidates based on in vitro efficacy and toxicity signals - https://www.biorxiv.org/content/10.1101/693572v3
After testing for PK/PD and tolerability in vivo, the process led to Deep Genomics identifying a potential antisense oligonucleotide (ASO), DG12P1, to correct the exon skipping for ATP7B (M645R). With a significant portion of restoration needed in the liver, ASO delivery should not be a problem for Deep Genomics in Wilson Disease. The entire process from discovery to lead took around 18 months and now is going toward IND along with other programs in metabolic diseases and neurodegeneration.
What makes Deep Genomics unique?
Deep Genomics combination of deep learning and ASOs gives them the potential to develop informational drugs. Which means getting to a point where a data scientist substitutes the role of a medicinal chemist or an optimization step.
The company’s AI Workbench platform focuses on 4 features:
Target identification - searching for alterations in splicing, gene expression whether transcription or translation, and protein activity
Lead identification - optimizing for on-target effects and minimizing off-target effects
ADMET prediction - relying on animal models and cell viability data
Trial design and endpoints - developing new biomarkers
The first step is finding a viable target for a given disease. This is pretty hard given the vastness of the universe for drug targets - picking one out of many that will have a major impact on patients is important. Ultimately, Deep Genomics has a shot to realize the potential of deep learning in drug development.
The other part of the business is also realizing the potential of ASOs. The 2 parts of Deep Genomics - deep learning and ASOs - are interlinked: deep learning can generate a large number of targets but need a modality that can predictable target them and ASOs have standard parts but need a system that can generate enough addressable targets. Deep Genomics will need to make investments in ASO technology to pursue more diseases. They can also use small molecules, biologics, among other modalities, which negate some of the benefits of their platform.
As background, ASOs are synthetic single-stranded nucleic acids around 18 to 30 nucleotides to control gene expression of specific targets. Gene expression can be controlled through complete silencing, splice modulation, and even activation. ASOs interact with their corresponding targets through complementary Watson–Crick base pairing. This mechanism for ASOs is pretty powerful: leads are generated pretty easily once the sequence of the target is identified.
Upfront investments in the chemical modification of the ASO backbone to improve their ability to enter a cell and remain stable and its delivery (i.e. nanoparticles, conjugation) are required but can be amortized over the use of the ASO:
More rapid discovery - just need a target and a biological hypothesis. There is some medchem focused on 3 parts: nucleotides, sugar backbone, and phosphodiester bonds.
MoA - just need a target to get started in drug development
Clinical development - one NME with multiple actions; new modality so regulatory controls still being mapped out
Manufacturing - standardized versus other modalities
Synthesizing all this together, once an ASO and its backbone chemistry is designed, the sequence it targets is easily interchangeable. So the role of bioinformaticians and data scientists in ASOs is equivalent to the role medicinal chemists have in small molecule discovery. This is where Deep Genomics shines giving them an opportunity to build a business model similar to Millennium Pharmaceuticals.
Why I like what Deep Genomics is doing?
With its platform, Deep Genomics has the potential to build a scalable drug development business model. Their internal pipeline has potential and it will be exciting to see their Wilson Disease program move into a phase 1 trial.
Moreover, the company has the ability to execute a partnership business model to power more medicines and generate high amounts of non-dilutive capital. Deep Genomics’ recent partnership with BioMarin to develop new ASOs in 4 rare disease indications is an early signal of this potential.
You can find Deep Genomics here.