Technology
Our core technology answers a fundamental set of problems in drug development and healthcare from first-principles thinking: we need to accurately model the dynamics of the system in order to predict the properties of the components within the system.
We are evolving fundamental methods to create a dynamical systems simulator for biology to provide all patients with more effective, safer medicines
We believe a systems biology approach is a necessary step towards ML and technology solutions that will result in more affordable, effective medicines. Our team is advancing fundamental methods in Geometric Deep Learning and Neural Graph Differential Equations to create a generalizable, dynamical systems simulator for biology.
The human biological system presents significant challenges of dynamic complexity and vast scale
We capture the dynamics and state-space of a very large-scale representation of the human biological system with 2,000,000 components or nodes per cell for around 2,000 cell types. Our work involves edge-of-chaos dynamics, multiple-attractor states, stability and control mechanics and more, all at large computational scale with models scaling to hundreds of millions of parameters.
This is a problem space only recently made tractable by advances in machine learning and high performance computing
Fundamental improvements in machine learning methods are transformational once they’re applied at scale (think Google Maps’ routing algorithm, or the Spotify recommendation engine). Open source and academic projects play a crucial role in advancing the field. Syntensor is a contributor and supporter of DiffeqML/Torchdyn, the leading open source library in the Neural Graph Differential Equations space, and we are the first to apply NDEs at massive scale.
Our methods mean drug developers, clinicians and physicians using our platform can ask, “Why?”
Our platform couples mechanistic, causal inference with an apps and services layer, enabling drug developers, clinicians and physicians to self-serve ‘hypotheses’: multi-scale inferences describing the perturbed state of a healthy cellular component, cellular process, tissue or patient when disease or drug is present, as well as the mechanism of action and ADMET properties of the drug itself.
This is not a ‘black box’ tool or statistical model. We predict the dynamics of the individual components of the system and the systemic effects of perturbation of the system, which means we can then capture the emerging phenotypic properties of the system when it is perturbed. Platform users can analyze the dynamic mechanistic drivers of the hypothesis; they can ask not only Will it work? but also Why?
