We envisage a world where therapeutics are side-effect free
Drug development is broken
Proportion of FDA approved drugs with at least 10 toxic
Cost to develop a new, probably toxic drug
Time to develop a new, probably toxic drug
We are guided by three tenets that will allow us to fix drug development
Drug development is a garbage in, garbage out model. If we start with a bad target, our ability to discover high quality drugs is fundamentally limited.
This is true for all stages of the drug development pipeline, from virtual drug screening through to animal and human models.
Current failures in drug development stem from poor models of disease at the very start of the pipeline.
2: AI Is Essential
There are 10 possible drug compounds, that's more than the number of stars in the universe. With this, we have massive untapped potential for treating disease. The challenge is finding the subset of drugs that are safe and effective.
AI is essential for mining the drug universe. With AI we can probe drug-target interactions at high accuracy using native representations of both chemical and molecular structure and function. This allows for rapid, prioritized searching of the chemical space.
Discovering safe, effective therapies relies on having data that accurately represents how drugs interact with their target disease inside the body. AI alone cannot extrapolate the complexity of chemical interactions from simple information about the drug and target.
We need to fuel our AI with high quality experimental data that accurately represents how drugs interact with disease, only then can we build powerful AI.
3: AI Is Insufficient
By trawling through databases and literature on drug-target interactions, we've built a proprietary knowledge graph: TOME, that maps drug-target interactions, accounting for detailed information about how targets behave inside the body.
Through TOME, we've identified significant gaps in data on drug-target interactions that ignores how targets behave inside the body. We're filling these gaps with advanced drug binding experiments to create a comprehensive database.
With comprehensive data that accounts for how targets behave in the body, we can build powerful AI based on deep representation learning for discovering new therapeutics that are safer and more effective for patients.
As we grow, we can continuously generate and integrate new data into our AI models. Starting with high fidelity molecular interactions, we can leverage genomics, proteomics, and phenomics to create complete models of drug-disease interaction.