Putting AI predictions into practice
An expert interview with Dr. Peter Staller, Head of Functional Genomics Research at Nuvisan Pharma Services, Dr. Filippos Klironomos, bioinformatician and data scientist at Nuvisan Pharma Service, and Dr. Thomas Wollmann, CTO at Merantix Momentum. Moderated by Dr. Gillian Hertlein, Strategic Project Manager at Merantix Momentum, the conversation will revolve around the use of Artificial Intelligence in the validation of pharmaceutical targets.
Gillian: Welcome to all of you! To begin, not all of our readers may be familiar with the concept of target validation. Peter, could you give us a brief introduction to what target validation entails and why it's essential for drug development?
Peter: Target validation is the crucial step that follows target identification in the pharmaceutical value chain. It aims to ensure that engaging a specific target, whether it's a protein, RNA, or another cellular component, using various modalities like small molecules or antibodies would have a therapeutic benefit for a particular disease. In other words: Target validation serves to confirm that the chosen target is a viable candidate for treating the disease. Essentially, it involves a deep characterization of potential targets that have been identified in earlier stages to provide a solid scientific rationale for pursuing the most promising one. This rationale is pivotal as it triggers a significant economic and time investment in drug discovery.
Gillian: You mentioned that target validation can vary depending on the type of target and disease. Could you give three examples of common approaches used to validate a target?
Peter: Target validation in pharmaceuticals heavily relies on model systems. These systems include in vitro models, where cellular systems are cultured in a dish, in vivo models using animal strains that mimic human diseases, and experimental tools to modulate the target in these systems. These tools could be small molecules or antibodies. In general, tool compounds are unsuitable for human use but can still validate the target in experimental model systems. In cases where we lack tool compounds or like to obtain complementary evidence, functional genomics techniques(1), like RNA interference(2) or CRISPR-Cas9(3), can help modulate target activity. Finally, genetically modified mice can be used, though this approach is time-consuming, complex, and not always possible due to the genetic differences between mice and men.
Gillian: Let's talk about whether simulation or machine learning can play a role in target validation. Filippos, do you think simulating the effects or generating data and applying machine learning to it can help?
Filippos: Well, I have a background in theoretical physics, so I've worked extensively with simulations. Simulating the molecular biology of single cells or the interaction of several different cell types is incredibly complex. There are currently no accurate models that can capture such complexity, and there is no data to train AI models to replace simulations in that task. We must rely on experiments to probe biology. To reliably target-validate, one needs models that capture the biology of disease. Therefore, one needs to add more complexity going from single cells to tissues, to organs, to the whole organism. For these reasons, I would say reliable target validation will remain under the domain of wet lab experimentation for some time. However, reliable target identification can become AI-dominated in the foreseeable future because, coming from that direction, even single-cell assays might be enough to capture causal drivers.
Gillian: Thank you, Filippos. Thomas, what are your thoughts on using machine learning in this context?
Thomas: It depends on where in the data analysis chain we apply simulations. When dealing with specific mechanisms within cells, we can build models. Machine learning can be useful for analyzing high-dimensional data and can extend classical statistical methods when working with sequence data. It can also help in integrating and analyzing multi-modal data, which can be valuable for understanding complex biological processes. However, simulating entire diseases or targets for validation purposes remains challenging, and I believe experimental validation is crucial.
Gillian: Now, turning to the use of large language models, how do you see their role in the future of target validation, especially in combination with AI and machine learning?
Peter: Large language models can act as world models, recognizing patterns from vast amounts of data. They can be useful for generative tasks, such as predicting sequences or for classifying tasks. In particular, they excel in handling and summarizing extensive knowledge, which can be highly valuable in curating information from scientific literature and databases. They can help researchers make sense of the vast web of knowledge in the field.
Filippos: Indeed, they can be a powerful tool in information retrieval and summarization. However, the choice of tool should align with the specific question at hand. For tasks like target validation, where experimental validation is paramount, large language models may play a supporting role in information gathering and analysis.
Gillian: What do you envision as the future of target validation? Where do you see the greatest potential for advancements, whether AI-related or not?
Peter: From my perspective, I see advancements in experimental model systems as a key to the future of target validation. Techniques like induced pluripotent stem cells (iPSCs)(4) hold tremendous promise. These cells allow us to create various human tissues in vitro, offering more complex and better translatable model systems. Additionally, improving image analysis, especially in patient samples, can provide insights into the molecular basis of diseases. AI can significantly aid in this area, helping to uncover crucial patterns and interactions.
Thomas: I agree that advancements in model systems are crucial. Furthermore, automation and machine learning can enhance the quality and objectivity of data analysis, especially in animal studies and experimental setups. However, we should always choose the right tool for the job, and AI can excel in knowledge curation and integration, making it a valuable asset in drug discovery.
Filippos: While I see great potential in AI for information retrieval and integration, I believe single-cell sequencing, coupled with spatial transcriptomics, will be a game-changer in understanding molecular biology and disease mechanisms. These technologies offer a holistic view of cellular interactions and gene expression, potentially revolutionizing target identification and validation.
Gillian: Thank you all so much for your valuable insights and perspectives!
__
(1)Functional genomics techniques are a set of molecular biology methods used to study how genes function within an organism's genome. These approaches involve manipulating and analyzing genes to understand their roles in various biological processes, often employing tools like RNA interference and CRISPR-Cas9.
(2)RNA interference (RNAi) is a molecular mechanism in cells that regulates gene expression by suppressing the translation or degradation of specific messenger RNA (mRNA) molecules. Researchers use RNAi techniques to selectively silence or reduce the expression of target genes, providing valuable insights into gene function and potential therapeutic applications.
(3)CRISPR-Cas9 is a revolutionary genome editing technology that allows precise modification of DNA sequences within an organism's genome. By utilizing a Cas9 protein guided by RNA molecules, scientists can edit genes with high accuracy, making it a powerful tool for genetic research, disease treatment, and the development of genetically modified organisms.
(4)Induced pluripotent stem cells (iPSCs)are specialized cells created by reprogramming adult cells, such as skin cells, into a pluripotent state, similar to embryonic stem cells. iPSCs have the capacity to differentiate into various cell types in the body, offering great potential for regenerative medicine, disease modeling, and drug development.
Subscribe now to the Merantix Momentum Newsletter.
More articles
Putting AI predictions into practice
An expert interview with Dr. Peter Staller, Head of Functional Genomics Research at Nuvisan Pharma Services, Dr. Filippos Klironomos, bioinformatician and data scientist at Nuvisan Pharma Service, and Dr. Thomas Wollmann, CTO at Merantix Momentum. Moderated by Dr. Gillian Hertlein, Strategic Project Manager at Merantix Momentum, the conversation will revolve around the use of Artificial Intelligence in the validation of pharmaceutical targets.
Gillian: Welcome to all of you! To begin, not all of our readers may be familiar with the concept of target validation. Peter, could you give us a brief introduction to what target validation entails and why it's essential for drug development?
Peter: Target validation is the crucial step that follows target identification in the pharmaceutical value chain. It aims to ensure that engaging a specific target, whether it's a protein, RNA, or another cellular component, using various modalities like small molecules or antibodies would have a therapeutic benefit for a particular disease. In other words: Target validation serves to confirm that the chosen target is a viable candidate for treating the disease. Essentially, it involves a deep characterization of potential targets that have been identified in earlier stages to provide a solid scientific rationale for pursuing the most promising one. This rationale is pivotal as it triggers a significant economic and time investment in drug discovery.
Gillian: You mentioned that target validation can vary depending on the type of target and disease. Could you give three examples of common approaches used to validate a target?
Peter: Target validation in pharmaceuticals heavily relies on model systems. These systems include in vitro models, where cellular systems are cultured in a dish, in vivo models using animal strains that mimic human diseases, and experimental tools to modulate the target in these systems. These tools could be small molecules or antibodies. In general, tool compounds are unsuitable for human use but can still validate the target in experimental model systems. In cases where we lack tool compounds or like to obtain complementary evidence, functional genomics techniques(1), like RNA interference(2) or CRISPR-Cas9(3), can help modulate target activity. Finally, genetically modified mice can be used, though this approach is time-consuming, complex, and not always possible due to the genetic differences between mice and men.
Gillian: Let's talk about whether simulation or machine learning can play a role in target validation. Filippos, do you think simulating the effects or generating data and applying machine learning to it can help?
Filippos: Well, I have a background in theoretical physics, so I've worked extensively with simulations. Simulating the molecular biology of single cells or the interaction of several different cell types is incredibly complex. There are currently no accurate models that can capture such complexity, and there is no data to train AI models to replace simulations in that task. We must rely on experiments to probe biology. To reliably target-validate, one needs models that capture the biology of disease. Therefore, one needs to add more complexity going from single cells to tissues, to organs, to the whole organism. For these reasons, I would say reliable target validation will remain under the domain of wet lab experimentation for some time. However, reliable target identification can become AI-dominated in the foreseeable future because, coming from that direction, even single-cell assays might be enough to capture causal drivers.
Gillian: Thank you, Filippos. Thomas, what are your thoughts on using machine learning in this context?
Thomas: It depends on where in the data analysis chain we apply simulations. When dealing with specific mechanisms within cells, we can build models. Machine learning can be useful for analyzing high-dimensional data and can extend classical statistical methods when working with sequence data. It can also help in integrating and analyzing multi-modal data, which can be valuable for understanding complex biological processes. However, simulating entire diseases or targets for validation purposes remains challenging, and I believe experimental validation is crucial.
Gillian: Now, turning to the use of large language models, how do you see their role in the future of target validation, especially in combination with AI and machine learning?
Peter: Large language models can act as world models, recognizing patterns from vast amounts of data. They can be useful for generative tasks, such as predicting sequences or for classifying tasks. In particular, they excel in handling and summarizing extensive knowledge, which can be highly valuable in curating information from scientific literature and databases. They can help researchers make sense of the vast web of knowledge in the field.
Filippos: Indeed, they can be a powerful tool in information retrieval and summarization. However, the choice of tool should align with the specific question at hand. For tasks like target validation, where experimental validation is paramount, large language models may play a supporting role in information gathering and analysis.
Gillian: What do you envision as the future of target validation? Where do you see the greatest potential for advancements, whether AI-related or not?
Peter: From my perspective, I see advancements in experimental model systems as a key to the future of target validation. Techniques like induced pluripotent stem cells (iPSCs)(4) hold tremendous promise. These cells allow us to create various human tissues in vitro, offering more complex and better translatable model systems. Additionally, improving image analysis, especially in patient samples, can provide insights into the molecular basis of diseases. AI can significantly aid in this area, helping to uncover crucial patterns and interactions.
Thomas: I agree that advancements in model systems are crucial. Furthermore, automation and machine learning can enhance the quality and objectivity of data analysis, especially in animal studies and experimental setups. However, we should always choose the right tool for the job, and AI can excel in knowledge curation and integration, making it a valuable asset in drug discovery.
Filippos: While I see great potential in AI for information retrieval and integration, I believe single-cell sequencing, coupled with spatial transcriptomics, will be a game-changer in understanding molecular biology and disease mechanisms. These technologies offer a holistic view of cellular interactions and gene expression, potentially revolutionizing target identification and validation.
Gillian: Thank you all so much for your valuable insights and perspectives!
__
(1)Functional genomics techniques are a set of molecular biology methods used to study how genes function within an organism's genome. These approaches involve manipulating and analyzing genes to understand their roles in various biological processes, often employing tools like RNA interference and CRISPR-Cas9.
(2)RNA interference (RNAi) is a molecular mechanism in cells that regulates gene expression by suppressing the translation or degradation of specific messenger RNA (mRNA) molecules. Researchers use RNAi techniques to selectively silence or reduce the expression of target genes, providing valuable insights into gene function and potential therapeutic applications.
(3)CRISPR-Cas9 is a revolutionary genome editing technology that allows precise modification of DNA sequences within an organism's genome. By utilizing a Cas9 protein guided by RNA molecules, scientists can edit genes with high accuracy, making it a powerful tool for genetic research, disease treatment, and the development of genetically modified organisms.
(4)Induced pluripotent stem cells (iPSCs)are specialized cells created by reprogramming adult cells, such as skin cells, into a pluripotent state, similar to embryonic stem cells. iPSCs have the capacity to differentiate into various cell types in the body, offering great potential for regenerative medicine, disease modeling, and drug development.