Cancer is a leading cause of mortality, accounting for nearly one in six deaths worldwide. Over 18 million new cases of cancer are diagnosed globally every year. To fight this battle, over $50 billion are spent on R&D in oncology per year. However, because of tumor complexity and diversity, the development of precise drugs targeting cancer is very challenging. That is why there are hundreds of cancer therapies currently in use and thousands more in development.
The efficacy of even the best therapies is far from perfect, while the success rate of putting a new drug on the market is very low (3.4%) with costs reaching $2,8 billion. Despite all this effort, most patients still don’t get the therapies they need. We aim to change this. Our mission is to provide the right therapy for every cancer patient. And this is done with the use of artificial intelligence.
Recent breakthroughs in artificial intelligence encourage experts from all fields to tap into the power of AI. These breakthroughs have been made possible by large-scale, high-performance computing, combined with enormous amounts of data. Cancer is a disease of the genome and thanks to the advancements in DNA sequencing, we now have access to genomic data from real patients.
However, the currently hyped AI approaches do not work in genomics because they rely on a simple premise: the more data, the better the model. For example, the large language models, capable of holding a conversation, reciting poetry, or generating computer code, have been trained on billions of pages of text, each containing a relatively small amount of information.
Genomics is different. Here, the datasets often consist of hundreds of samples with billions of nucleotides describing each one of them. It is seemingly impossible to distinguish any signal from the noise with this much information to consider. But it isn't. We have done it, we have validated it externally, and we have developed a technology allowing us to do it at scale. Our solution is precision AI.
Precision AI is a fusion of machine learning and biological knowledge. We have developed a platform designed to infuse machine learning models with biological knowledge allowing the models to focus only on the relevant parts of the genome. We process each genome using proprietary pipelines which inject more and more biological knowledge at each step. As a result, each step acts as a sieve filtering more and more noise from the data until only the relevant signal remains. This allows us to extract increasingly complex biological concepts from the genomes and combine them together to construct a model of tumor biology. Not only does our model get better with more data, but most importantly, it improves as we add on more biological knowledge.
The model helps us truly understand the processes taking place in cancer and lets us verify various hypotheses regarding both existing and potential therapies.
First, we are able to identify patients who will respond to known therapies, boosting the effective efficacy of the drugs which are already in use or are entering clinical trials. Second, we can observe cohorts of patients who would benefit from different therapies than those they were originally assigned to, finding other indications for existing drugs and facilitating drug repurposing. Finally, the model shows us cohorts of patients for whom there are no right therapies available, allowing us to pursue and validate new drug target candidates.
All of this is possible thanks to our model of tumor biology developed using our precision AI approach. This is how we fulfill our mission. This is how we find the right therapy for every cancer patient.