By Noora Kattouf.
By now we are all familiar with the potential applications that generative AI has across different industries and especially the many promises of optimizing and accelerating drug discovery processes in the pharmaceutical industry. Understanding the natural evolution of drug discovery is imperative for making the best future investments. We asked Dr. Kashif Sadiq, CEO and founder of DenovAI (an AMITI portfolio company) to share with us his thoughts on the future of medicine and human health. Here is our summary:
What are the challenges that the pharmaceutical industry faces that can be solved with AI/ML? And once we've tackled these challenges, what kind of future do we envision?
Kashif: Developing a drug can take 10-15 years and cost a pharmaceutical company $4-10B. The drug development process involves multiple stages, such as discovery, pre-clinical testing, and clinical trials, and each of these stages is filled with the potential for failure. The promise of AI is to remove unnecessary steps, not only the ones that are not needed but also those that take too much time and could be more effective.
For example, discovering new antibodies is a complicated process involving multiple antibody sequences. These sequences are used to identify potential hits. However, the experiments are not only time-consuming but also quite expensive. Suppose you bypass the current discovery process with an in-silico solution. In that case, you will still need to validate the molecules experimentally to ensure patient safety and efficacy. However, this method saves time and cost by eliminating the majority of experiments and validating only at the end, making the process much more efficient and effective.
That is only part of the development pipeline; there are other parts to the development process, and AI can play a role in every part of this pipeline, which is why it is so appealing.
What makes proteins so essential to our bodies, and why is it important to understand how they interact with each other to develop new medicines?
Kashif: Proteins are essential to all living things. At the molecular level, the most critical factor that governs all life is the capability of molecules to bind to one another. Thus, all processes in a functioning cell and the signals that pass between cells depend on proteins finding and binding to each other long enough to transfer the signal and initiate a biochemical reaction within the cell. These proteins then detach and bind to other proteins, relaying the signal in complex networks of protein-protein interactions.
Understanding protein-protein interactions is crucial from both a scientific and medical perspective. Many diseases are caused by proteins or protein interaction networks that misfunction, so it's essential to identify where these interactions go wrong; by doing so, we can design proteins and other therapeutics that can bind to the misfunctioning proteins and interfere with their activity. We analyze proteins at their atomistic level. By examining the thousands of atoms that make up a protein, we can understand not only how the protein interacts with itself and its surroundings but also with other proteins and molecules with which it can bind. This helps us predict how strongly and specifically a particular protein will bind to its partners, similar to a lock and key mechanism.
If we achieve that, it gives us fantastic abilities to create new medicines and interfere precisely in interrupting those protein-protein interactions that then lead to disease.
At Amiti, we always invest in the best team for the mission, but when investing in generative AI our thesis is to also invest in the best data and best approach for the mission. Why is AI alone not enough and how is DenovAI different from existing and emerging companies?
Kashif: Generative AI is a revolutionary technology that creates latent spaces to generate new creative solutions. These solutions are made possible by the advancements in AI architecture, particularly the transformer technology. Chat GPT and large language models are some of the products of this technology. On the biology front, it has also led to the development of protein language models, which allow users to input a protein sequence and generate the rest of the protein automatically. Generative AI methods used in protein design are based on this principle.
However, there is a caveat to this approach. Despite Generative AI's impressive capabilities, it is insufficient on its own. DenovAI believes that "AI proposes, and physics disposes". While AI can propose numerous solutions, most of them are likely to be wrong.
Imagine using generative AI to create lions and lion-like things. You'd produce millions of iterations, each with small variations. With millions of iterations, each with slight variations, some may feature toes instead of knees or vice versa. As humans, we know that a functioning lion with toes in place of its knees is impossible, as this goes against the laws of physics – the lion wouldn’t be able to run. However, analogously, it can be challenging for humans to discern the proper placement of "knees and toes" in molecules. Therefore, using physics to determine when and why AI goes wrong is imperative. This methodology is what DenovAI does differently; it generates numerous possibilities, which are then tested against physical models to determine their feasibility. This approach provides a realistic ground reality that no one else to our knowledge is currently doing.
The second part is about data. It's incredibly important, and everybody knows that. One of the significant benefits of the AION Labs model (where AMITI is a board member) is that it enables access to publicly unavailable data and within the pharma partner's domain to develop the AI model, making it incredibly powerful. DenovAI can fully use the data while ensuring the privacy of the individual pharma partners who do not want their data shared amongst themselves. Technology solutions already enable such federated data sharing, respecting privacy while allowing us as a whole to be more than the sum of our parts, which is essential.
What do you think the future of medicine will look like in a decade from now. What other challenges in human health can we overcome with protein design?
Kashif: What we discussed earlier was just one aspect of disease: the design of antibodies and proteins for therapeutics in traditional methods. However, other exciting healthcare technologies exist, such as vaccines and regenerative medicine, that hold the promise to regenerate tissues and organs. Protein design can come to the call of that; by understanding the signaling mechanisms of the body, it could be possible to have proteins and antibodies designed to control the disposition of where the cells go, how they form and how they differentiate into different types of cells and that will one day lead to the ability to generate new organs.
Another important area is pandemic preparedness. Rather than waiting for a pandemic to strike and developing a treatment in response, it is better to proactively create an in-silico therapeutic. With the help of in-silico therapy, we can be better prepared for possible emerging pandemics. Doing so allows us to produce the necessary medication to treat any potential outbreak quickly. So, instead of waiting for a pandemic to hit us, we can take action now and potentially save countless lives.
Going beyond medicine and human health - you once mentioned that having mastery over protein design leads to mastery over multiple domains in human existence. Tell us what does that mean, how can we solve bigger climate challenges with protein design?
Kashif: The possibilities are endless; there are huge areas where protein design could benefit. The energy sector of renewable energy relies on electrolytes, with proteins playing a key role in solar cells. Another critical area of concern is agricultural feed and food crops. With the world's population on the rise, we will need help generating enough food. Therefore, increasing crop yield and agricultural feed uptake will be crucial in boosting food production. We must tackle this huge challenge to ensure a sustainable future for our planet. Industrial enzymes that we can design will significantly impact various fields, including sustainable biomaterials. A lot of other biomaterials made directly from proteins could be synthetically produced for clothing. Why only silk? This is why at DenovAI we see huge opportunities as well in synthetic biology across the supply chain of multiple sectors and not just in pharmaceutical development.
Kashif, thank you for sharing these great insights with us.
At AMITI, we are very excited about transformational tech that will change the world for the better and we see a very positive future possible with the technologies that our companies are building.
