Molecular motion, accelerated
Life Is Movement
Inside every cell, proteins orchestrate the processes essential to life. Their interactions, choreographed over billions of years of evolution, are responsible for catalyzing reactions, transmitting information, and keeping our cellular processes healthy. But these dynamic processes can falter and slip into states we recognize as disease. Therapeutic intervention seeks to restore the natural order, and to do that effectively we need to see and understand what proteins do.
Many proteins fold into relatively stable three-dimensional structures. Others, known as intrinsically disordered proteins, lack a stable principal structure and instead continuously shift among many different shapes. Successful therapeutic intervention requires understanding the range of shapes they adopt and how they move between them.
At Topos Bio, we use molecular dynamics simulations and new AI models to explore how proteins move and interact with their surroundings. Combined with experimental observations, these computational approaches are revealing how molecular motion gives rise to biological function and dysfunction, creating new avenues for therapeutic intervention.
Catching Up to Nature’s Pace
Proteins move at astonishing speed. Vision begins with a molecular movement on the order of picoseconds, roughly 200× faster than a 5 GHz CPU clock cycle. Simulating one microsecond of a protein’s motion can occupy a powerful GPU for a week on the systems we study at Topos.
We’re working with NVIDIA to bring our computational capabilities closer to nature’s pace. Topos has already generated molecular simulations for over 100,000 different systems, with well over one trillion time steps evaluating all the forces and interactions a protein makes with itself and its surroundings. Each one takes us a step closer to our goal of predicting the motions relevant to treating illness.
Working closely with NVIDIA Principal Product Manager Yutong Zhao, our team implemented NVIDIA’s cuPotential technology into our own OpenMM simulation workflows. These efforts increased molecular dynamics throughput by 2.4×, and in some workflows by up to 3.6×. These advances allow us to generate more molecular dynamics data to build an increasingly rich foundation of physical knowledge for training our generative AI models.
Topos Bio is taking even bigger strides with our growing family of generative AI models. Trained on our growing simulation library, Topos-1, Topos-2, and Topos-Bind learn molecular behaviors and explore broader regions of conformational space without calculating every detailed step along the way. These expanded landscapes, including rare conformations that conventional molecular dynamics simulations may never reach, offer new opportunities to influence protein function.
Moving Towards Medicines
At Topos, our experimental platforms bring these expanded molecular landscapes into contact with the physical world. In less than a year, we’ve screened more than 10 billion molecular interactions, measured over 10,000 time-resolved binding events, and collected hundreds of X-ray scattering measurements of proteins in solution. Together, these observations help reveal how flexible proteins behave and how they recognize other molecules.
Simulations help us interpret these observations while experimental evidence tests and refines our models. This growing understanding guides what we design next through our generative chemistry platforms. Topos-Chem designs new molecules, while Topos-AP ranks promising candidates for synthesis and testing. The results, in turn, guide the next round of discovery.
The connections we’re making between simulation, experiment, and the effects of our therapeutic molecules are opening new paths toward medicines that reshape what’s possible in treating disease.
We invite researchers and potential partners to join us in exploring how molecular motion can open new possibilities for treating disease.
Changing the dynamics of disease
Contributions
- Molecular Dynamics & cuPotential. Jiaqi Zhu.
- Blog post & visualizations. Anthony M. Giannetti.