A co-creative process between AI and humans

Welcome to the 4 new deep divers who joined us since last Wednesday. If you haven’t already, subscribe and join our community in receiving weekly AI insights, updates, and interviews with industry experts straight to your feed.

DeepDive

Your weekly immersion in AI. 

With a background in biotech and biophysics, Thras Karydis (Co-Founder and CTO at DeepCure Inc.) is part of the founding team behind DeepCure – working to accelerate breakthrough science and leverage AI to improve drug discovery for the healthcare sector. 

We asked him about the journey to developing DeepCure’s unique technology, and the ‘aha’ moments along the way. 

Could you share the story behind DeepCure?

“Kfir Schreiber, Joseph Jacobson and I started DeepCure at MIT in 2018 with the mission to overcome the toughest challenges in small molecule drug discovery. We recognised that traditional methods failed for many targets and believed in a very different approach to other companies using AI. 

“We shared the predictions from our early AI platform with a senior executive at a large pharmaceutical company and they were super impressed, which led to the founding of DeepCure. 

“Today, our innovative platform spans AI, physics-based methods and automation. The platform enables us to scan the protein surface for novel binding sites, design truly novel molecules that meet stringent requirements, and transform designs into testable molecules. 

“DeepCure is now a therapeutics company with a pipeline of assets for inflammation and immunology diseases, including one compound we expect to enter clinical trials next year and could transform the treatment of arthritis.”

When did you begin to recognise the potential of AI for drug discovery and patient care? 

“We believe in the potential for AI to improve drug discovery and healthcare. That belief was confirmed in one project where we sought to find potent, selective and brain penetrant molecules for a target. 

“Initially, we had no success when we searched a huge, commercially available library of 20 billion compounds. But when we allowed our AI to design compounds without limiting them to commercial libraries, we saw immediate success. Many of the AI molecules were potent, selective and/or predicted to be brain penetrant compounds. 

“This was a huge ‘aha moment’ for us. That’s when we realised that we (and others) were holding back AI by imposing too many restrictions that were tied to past biases and historical compounds.  

“We’ve now removed these biases and restrictions on AI that limit originality and focused on using it as a creative force for our chemists, which has led to exciting breakthroughs on many of our programs.”

What is unique about DeepCure's technology?

“The uniqueness of DeepCure’s technology begins with the integration of AI, physics, structural biology and chemistry to focus on novel solutions for hard problems. That’s true of both our team and our technology. Our approach starts with a unique proprietary approach (PocketExpander™) that combines AI, physics and structural biology to map previously uncharted protein surfaces for novel features using. 

“This approach allows us to pursue undrugged targets with no known ligands and to design challenging molecules such as molecular glues to stabilise protein-protein interactions, which are impossible or very challenging problems for other technologies. 

“We then use our MolGen™ tool, which combines AI, chemistry and physics to create custom libraries of novel compounds that are synthesizable and optimized. 

“These custom libraries enable us to explore compounds that are unlike known compounds and avoid bias towards past chemical compounds. These unique capabilities have allowed us to tackle drug discovery challenges that others can’t and succeed, and importantly we have built an interface that enables these technologies to be used by chemists and biologist researchers and for AI results to be interpretable.”

Why is it important for AI models in drug development to be explainable and interpretable by humans?

“Drug discovery and development is a truly interdisciplinary effort. No one person or tool can create every solution or address every challenge, and it makes no sense to hire a great team and ask them to set aside their knowledge, creativity and decision-making to blindly follow AI predictions. 

“Therefore we not only push our technology to generate truly novel and innovative predictions, we also require that the results be explainable and interpretable. Our AI predictions are independently evaluated using physics-based methods, such as quantum dynamics and molecular dynamics, and have created tools so that human experts can review, interpret, visualise and refine AI generated outputs. 

“This co-creative process that fuses human insight and AI we have found accelerates the process and results in greater rigour, and most importantly we believe it will lead to drugs with better safety and efficacy for patients.”

Ready for DeepFest?

Thanks to Thras Karydis at DeepCure. If you want to learn directly from Thras and a host of other leading AI experts, register now to attend DeepFest 2025.