follow us us on Twitter
Subscribe to our mailing list
by Sarah Banducci
Verseon co-founder and Vice President of R&D David Kita discusses the two computational engines powering Verseon’s drug discovery platform and how Verseon’s computer-driven process can change the way new drugs are developed. In a previous post he gave a look into the origins of the company’s platform. Here, he goes into more detail on the computational engines and the team propelling Verseon’s platform.
>> In your article you mentioned the two computational engines that underly Verseon’s platform. You already discussed that the first computational engine was designed to model molecules against target proteins. With so many possibilities, how do you know which small molecules to test?
DK: In principle, our molecular modeling engine can test any small molecule against any target protein on the computer. The only limitation is that a suitable 3D structure of the protein must be available. We built our molecule creation engine to provide fuel for this process. We don’t just assemble random atom combinations, which would most of the time result in molecules that are impossible to synthesize and unsuited for drug development. Instead, the MCE designs new molecules that are both drug-like and synthesizable.
>> How did the molecule creation engine come together?
DK: We essentially needed to teach the computer to accurately apply hundreds of diverse synthetic chemistry reactions. Our team of chemists and computational engineers worked together to represent a large synthetic chemistry knowledgebase in silico. We have also created a virtual library of approximately 100,000 distinct, synthesizable building blocks that the molecule creation engine attempts to match together using its knowledgebase. This process is what allows us to create a practically unlimited supply of novel, drug-like, and synthesizable, virtual compound designs.
>> Why not use artificial intelligence (AI) to predict outcomes?
DK: More and more companies are putting forward big claims of using AI in drug discovery and development. While there are promising applications of this technique, like determining if a specific drug will be toxic or if it has a promising pharmacokinetic profile, the truth is that AI, or more specifically machine learning, is not currently capable of uncovering novel drug candidates.
Fundamentally, AI is based on looking for patterns in sets of data. Molecular systems, however, are inherently governed by the laws of physics and not well-suited for pattern fitting. In addition, data on known molecules, by definition, do not include novel chemical structures—a severe limitation for training-based methods. As a result, machine learning algorithms often end up looking for patterns in multidimensional noise.
Over ten years ago when the AI hype started, we foresaw the limitations of these methods and stayed on our original course.
>> How are these computational engines integrated in Verseon’s overall drug discovery and development process?
DK: When we target a specific disease-causing protein, the molecule creation engine first generates tens of millions of small molecules. Then, the molecular modeling engine models them in combination with the protein and determines a set of the best binders. Each of these molecules represents novel chemical matter—molecules that likely haven’t been tested or even synthesized.
From this list of virtual small-molecule binders, our chemists pick the most promising ones to synthesize. Synthesized candidates are then vetted by a comprehensive laboratory workflow consisting of a panel of in vitro and in vivo tests to ensure they are viable drug candidates before we proceed to clinical trials. By doing so, we are able to develop multiple clinical candidates for each disease program unlike most other pharmaceutical companies.
>> What happens if the candidate doesn’t meet all requirements off the bat?
DK: In concert with the molecule creation and modeling engines, our medicinal chemists can make adjustments and explore fine-scale modifications of the chemical structure to better satisfy one or more biochemical assays. By doing so, they can preserve the good features of the original candidate, while making the necessary optimization. This workflow is unique to Verseon and allows us to efficiently and systematically develop new drug candidates.
David Kita is co-founder and Vice President of R&D at Verseon.
- drug discovery -