Even drug-resistant cancer targets can be tackled with the right design strategy 

Your work leverages AI and machine learning to optimise CAR-T cell therapy. How does this approach fundamentally differ from traditional drug discovery and development pathways? 

CAR-T therapy is a powerful treatment used to fight cancer, especially blood cancers like leukemia and lymphoma. It is a type of immunotherapy, which means it uses the patient’s own immune system to attack cancer cells. In this treatment, doctors take the patient’s T-cells (a type of immune cell) and modify them in the lab so they can better recognise and kill cancer cells. These modified cells are called CAR-T cells. They are designed with two main parts: one part helps them recognise and attach to cancer cells, and the other part activates the T-cells to destroy those cancer cells. 

Once these modified cells are given back to the patient, they can find and kill cancer cells more effectively. However, one major challenge is that cancer cells can change over time. These changes (mutations) can alter the markers (antigens) on their surface, making it harder for CAR-T cells to recognise and attack them. 

In our Computational Structural Biology laboratory, we are working on designing better CAR-T cells that can overcome this challenge. We use advanced tools like computational modeling, artificial intelligence, machine learning, and molecular dynamics simulations to design these receptors. Our goal is to create more stable and effective CAR-T cells based on energy and structural analysis. These designs can then be further tested and developed for future clinical studies. Our approach using machine learning and molecular modeling is quite different from traditional CAR-T therapy. Normally, developing a new therapy is a long and expensive process that involves a lot of trialand-error experiments in the lab. Scientists test many different options step by step, which can take years. 

In contrast, machine learning and molecular modeling allow us to speed up this process by using computers to predict which designs are most likely to work. Instead of physically testing thousands of possibilities, we can first screen and optimise them virtually. This helps us quickly identify the best CAR-T cell designs with higher accuracy. Another key difference is that traditional methods are often more general, while machine learning and molecular modeling based approaches can be more precise and personalised. We can design CAR-T cells that are better suited to specific cancer targets or even individual patient variations. 

Overall, our approach reduces time, cost, and effort, while improving the chances of developing more effective and targeted CAR-T therapies. 

With R&D costs rising and clinical failure rates remaining high, how can computational structural biology help de-risk earlystage drug development and improve success rates? 

With drug development becoming more expensive and many drugs failing in later stages, computational structural biology offers a smarter and faster way to reduce risk early in the process.

In simple terms, instead of relying only on trial-anderror in the lab, we use advanced computer tools to understand how drugs interact with their targets (like tubulin) at a very detailed level. This allows us to predict which drug candidates are most likely to work-even before testing them in the lab. 

For example, in our work on cancer targets like tubulin and CAR-T cell design, we use methods like molecular modeling, simulations, and machine learning to: 

• Identify the best drug candidates early 
• Understand why some drugs fail (such as due to mutations or weak binding) 
• Design better molecules that can bind more strongly and effectively

This helps in “de-risking” drug development because: Fewer weak or ineffective candidates move forward to expensive lab and clinical testing.

Potential problems (like drug resistance or poor binding) are identified early Time, cost, and effort are significantly reduced While laboratory and clinical testing are still essential, computational approaches act as a powerful filter at the beginning. They improve the chances that only the most promising and safer drug candidates move ahead. 

Overall, computational structural biology makes drug discovery more efficient, more targeted, and more likely to succeedhelping bring better treatments to patients faster.

You’ve used Molecular Dynamics simulations to study drug resistance in cancer cells, particularly with drugs like Taxol. What key insights have emerged from this research, and how can they influence future oncology treatments? 

In our earlier research, using computer based molecular modeling and molecular dynamics (MD) simulations helped us understand drug resistance in cancer at a very detailed, molecular level – something that is difficult to observe through experiments alone. One key insight from our study is that mutations in the tubulin protein (the target of drugs like Taxol, Vinblastine and Colchicine) can change how the drug binds. Even small changes in the protein structure can reduce the binding strength of the drug, making it less effective. MD simulations showed that these mutations can alter the flexibility and stability of important regions in the protein, especially near the drug-binding site. Another important finding is that drug binding is not static-it is dynamic. The interaction between traditional drugs such as taxol, vinblastine and combretastatin, and tubulin changes over time, and resistant mutations can disturb this interaction, leading to weaker or unstable binding. This explains why some cancer cells initially respond to treatment but later become resistant. 

We also observed that changes in microtubule behavior and dynamics contribute to resistance. Cancer cells can adapt by altering these dynamics, reducing the effectiveness of drugs designed to stabilise them. These insights are very important for future cancer treatment because: 

• They help us understand why drug resistance occurs at the molecular level. 
• They enable the design of new drugs that can bind more strongly, even in mutated proteins. 
• They support the development of personalised therapies based on specific mutations in patients. 

Overall, our work shows that machine learning, molecular modeling and MD simulations can guide the design of next-generation anti-cancer drugs that are more effective and less prone to resistance, improving long-term treatment outcomes. 

Your research claims to significantly reduce drug discovery timelines from years to months. What are the practical challenges in translating these computational findings into clinical and commercial applications? 

As mentioned earlier, cancer can become resistant to treatment because the proteins in cancer cells change over time. These changes make existing drugs less effective, so there is a need to find new drugs that can still work against these mutated proteins. To tackle this, we use natural compound libraries and advanced computer methods to find molecules that can bind well to these mutated proteins and correct their behavior. We combine computer-based drug design with machine learning to quickly identify and improve the most promising drug candidates. Once we identify these compounds, we test them in the laboratory using cell-based experiments and by studying how strongly and effectively the drug interacts with the target protein. 

Although this approach can reduce drug discovery time from years to months, there are still several realworld challenges: 

• All computer predictions must be tested in the lab and in living systems, which takes time and resources. 
• The human body is very complex, so results from computer simulations may not always work the same way in real life. 
• Even if a compound works well in theory, it must be proven safe and non-toxic for humans. 
• Machine learning depends on good quality data, and poor or limited data can affect accuracy. ● Finally, before reaching patients, every drug must go through strict clinical trials and regulatory approvals, which are time-consuming and expensive. 

Overall, while computational methods help us move faster and save time, cost, and effort in the early stages, careful testing and approvals are still essential to ensure the treatment is safe and effective for patients. 

AI-driven redesign of immune cells is a significant advancement in immunotherapy. Can you elaborate on how your work enhances the precision and efficacy of CAR-T cells, especially in targeting markers like CD20? 

We use advanced computer tools, such as machine learning, molecular modeling and molecular dynamics simulations, to carefully design the part of CAR-T cells that recognises cancer markers on B cells, such as CD20, BCMA, and CD19. Our goal is to improve how precisely these engineered immune cells can identify and bind to attack cancer cell markers. 

A key focus of our work is improving how strongly and accurately CAR-T cells bind to cancer cell marker. To do this, we design different versions of the binding region (called scFv), based on antibodies using machine learning and molecular dynamics simulations followed by energy calculations. This allows us to predict which designs are likely to work best before moving to laboratory testing.

One important finding from our study is that even small design changes-such as modifying the linker that connects different parts of the binding region-can significantly improve the strength and stability of binding. This is especially important for targets like CD20, which are located very close to the cancer cell surface and are harder to access. 

Overall, our approach helps create CAR-T cells that are more precise in targeting cancer cells and more effective in killing them. By improving their stability and performance, this work has the potential to make CAR-T therapy more reliable and beneficial for patients with cancer. 

You are also exploring targets like tubulin for drug-resistant cancers. How promising is this approach in addressing socalled ‘undruggable’ or treatment-resistant malignancies? 

Targeting tubulin is still a very promising way to treat cancers, even those that have become difficult to treat due to drug resistance. From our research, we found that tubulin proteins in cancer cells are not all the same. There are different types (called isotypes like ßI, ßII, ßIII, etc.), and these differences can make common cancer drugs like Taxol and vinblastine less effective. This is one of the main reasons why some cancers stop responding to treatment. 

Using advanced computer methods, we studied how these drugs interact with different tubulin types. We found that drug resistance happens because of small changes in the structure and movement of the tubulin protein. These changes make it harder for drugs to attach properly and do their job. 

The good news is that this understanding helps us find solutions. By knowing exactly how these proteins change, we can design or identify new drug moleculesespecially from natural sources-that can bind better to these resistant forms. 

This machine learning, and molecular modeling approach is promising because: 

• It helps us understand why drugs stop working, 
• It allows us to design more targeted and effective treatments, 
• It opens the door for personalised therapies based on the patient’s cancer type Our work shows that even drug-resistant cancer targets can be tackled with the right design strategy. Overall, this approach can lead to better and more reliable treatments for drug-resistant cancers, making therapies more effective and less likely to fail. 

 

[email protected]
[email protected]