Transforming cancer treatment with the power
of artificial intelligence and network biology

The past decade saw the rise of systems biology. This discipline seeks to understand complex biological phenomena like cancer by describing them as networks of interconnected nodes. We are at a time when sufficient research has finally accumulated to accurately model cell signaling and other mechanisms as networks. However, static networks cannot capture the highly interlinked cascades of processes that start when introducing a therapeutic intervention.

Only network dynamics - turning networks on - can simulate the complex response of a tumor to therapies. But only small scale networks of a few dozen proteins ready for dynamic simulation have been published. Combining findings from thousands of published papers, we used network biology to create the first, high-definition virtual cell consisting of over 1100 proteins and their interactions.

Though this complex system can temporarily assume any state from untold billions of possibilities, over time they converge on a limited number of so-called attractor states. Research indicates that these attractors closely correspond to the observed phenotype of the given cells. In cancerous cells, proliferatory behavior dominates the attractor landscape.

Turbine finds treatments that change unwanted attractors, like proliferation in cancerous cells to more preferable ones, like apoptosis, finding ways to kill the diseased cells with novel combinations or applications of treatments. To do so, it runs millions of simulations to reveal the effects of all available therapies and their combinations on the system, feeding the results to an artificial intelligence. The AI then rapidly hones in on the optimal interventions.

Constructing the Simulated Cell

  • Turbine’s Simulated Cell is a directed and weighted signaling network, which incorporates 25 important signaling pathways acting in cancerous cells, including MAPK, ErbB, Notch and Hedgehog, describing all hallmarks of cancer.

Customizing the Simulated Cell to cancer type and patient

  • Turbine’s simulations closely resemble actual in vitro and in vivo interactions because we use available omics data to customize our cell model to a cell line or patient sample. Genetic mutations and transcriptomic information are both incorporated.

Compiling the simulation’s drug library

  • To find effective treatment combinations, available drugs and their targets are mapped to the simulated cell’s nodes. Compounds can include a client’s entire compound library, approved treatments for any disease, or other available public or in-house molecules. No structural data is required.

Modeling therapy response with the Turbine simulator

  • Turbine’s network simulator runs thousands of simulations on the customized cell to model the behavior of the network before and after therapies are administered. Turbine looks for so-called attractors, the steadiest states of the network that define how a cell functions or reacts to certain interventions.

Designing optimal therapies with artificial intelligence

  • As simulations are very fast, Turbine uses an artificial intelligence to select and discover the most effective therapy combinations among trillions of options. Our AI is a global optimization algorithm developed specifically to find effective combinations in the large search spaces of dynamic networks.

Publications from the Turbine team