Crash Course to Using AI to Determine Dose Selection

The development of a drug is a complex and lengthy process, requiring precise decisions about dosing to ensure both the efficacy and safety of a drug. Artificial intelligence (AI) can be used for dose selection in drug development—helping to save time, money, and increase accuracy. Before embarking upon this journey, it is important to grasp not only the basics of dose selection and why it’s important in drug development, but also how exactly AI improves the dose selection process. Then, once you understand these core concepts, you can begin evaluating various AI-based solutions for your specific needs. With AI on your side, analyzing and selecting dosage can become faster and more accurate than when determined by traditional methods.

What is Dose Selection & Why Is It Important?

Dose selection is a critical component of drug development, as it determines the amount of a drug needed to create an effective and safe response in patients. It’s important to understand the range of doses that are effective for a particular drug, and this range should be based on existing data from both animal and human studies. Equally as important as effectiveness is the safety and toxicity associated with dose selection, because too much or too little of a particular drug can lead to adverse effects or even death in some cases.  

The most difficult part of dose selection is often the “first-in-human” dose, because even when all animal studies are completed, researchers still will not know whether the dose tested in animals is safe for humans. Clinical trials traditionally begin with very low doses that are escalated gradually, and this process requires numerous clinical trials with many volunteers involved—the entire process is laden with risk of serious adverse effects. 

This is one area where AI can help define the safe first-in-human dose by estimating potential risks related to both on-target and off-target toxicity. In doing this, AI can help reduce the risk associated with human trials and help drug developers reach an answer about dose selection faster while also cutting down on the high costs of conducting unnecessary animal experiments and multiple clinical trials. 

Why is AI Winning Out Over Traditional Methods?

Traditional methods such as trial-and-error testing may still be employed in some cases of translational/animal science, but they are often too slow or costly when faced with complex problems that require precision and speed. Methods like trial-and-error are also unacceptable in the context of human studies, which often leads drug developers to perform extensive animal studies—for example, toxicity studies are mandatory to be carried out in two animal species, rodent and non-rodent, and are to last for six months minimum. Those reasons lead to a dramatic increase in drug development costs.

This is one reason that AI has become increasingly pursued for use in preclinical research. AI cannot help drug developers skip human or animal testing entirely, but it can help reduce risks of getting negative results (such as unexpected toxicity) as well as reduce the number of animals used by narrowing the dose range for the test drugs. By using AI algorithms such as natural language processing (NLP), machine learning (ML) techniques can be employed to identify patterns in data sets that would otherwise remain hidden from view—helping researchers make more informed decisions about dosing regimens based on real-time data. The understanding of these patterns is then used to predict therapeutic response across a range of dosages, revealing the optimal dose for humans - even across different patient subgroups when paired with sophisticated virtual patient stratification analysis. 

The technology offers numerous benefits past just increased accuracy and speed, such as being able to integrate information from multiple sources, and scale easily. These advantages can save time and money, making the decision-making process more efficient while providing an optimal dosage with fewer risks of errors. Ultimately AI can help to predict (so that one can avoid) a drug candidate’s likelihood to fail clinical trials in effectiveness or in safety. The latter area is often the trickier one, as it’s easier to determine an effective dosage of a drug than it is to find one that is both effective and safe. 

Furthermore, AI models are constantly updated with new information to keep up with the ever-evolving field of medicine. [Further reading: Barrett, J.S., Goyal, R.K., Gobburu, J. et al. An AI Approach to Generating MIDD Assets Across the Drug Development Continuum. AAPS J 25, 70 (2023). https://doi.org/10.1208/s12248-023-00838-x]

Considerations when Integrating AI into Dose Selection Research

When using machine learning to optimize first-in-human dosing for safety and effectiveness, you’ll need to consider several factors such as the type of data sets available (e.g., patient records, lab results), the parameters used in the model (e.g., age, sex, weight), and other variables that might influence the accuracy of your model (e.g., environmental conditions). Also important to specify are the therapeutic area and indication for drug use, including route of administration, amount of doses required, length of therapy, etc. Weigh up the advantages and disadvantages of using machine learning for this purpose—while it may be faster and more accurate than traditional methods, there are still potential issues, including regulatory requirements, that must be taken into consideration before moving forward with any project involving AI. Other facts to consider are computational cost, expertise of the personnel who will be using the platform, and scalability of your AI program overall.

Once you’ve considered all relevant factors and determined that machine learning is appropriate for your situation, you should then validate the accuracy of your model before deploying it as part of your drug development program. The FDA provides guidance on the topic of modeling and simulation in drug development, and part of this guidance requires that all models for drug development be validated and their accuracy confirmed before implementing those models for both preclinical and clinical studies. Doing so will help ensure that you can get reliable results from your model when selecting doses for patients or populations based on their individual physiology or other characteristics. It will also help to ensure the safety of the drug and help avoid unwanted effects of a toxic dosage.

Succeeding with AI

AI initiatives are often started within drug development organizations as proof-of-concept projects. Building an AI solution targeted at supporting preclinical translational research such as does selection is not trivial. AI models require significant training to achieve useful levels of accuracy in predicting efficacy/safety, and yet must also be designed for scalability beyond the initial proof of concept subject. Avoiding systemic bias in the AI system is a key requirement, as this can affect the accuracy of results, especially at scale. Furthermore, the level of expertise needed to deploy and maintain the AI system should be assessed before committing to a solution. By taking all these factors into consideration when employing AI for dose selection, companies can make sure that they are choosing the best fit for their specific needs.

VeriSIM Life offers an alternative approach. We deliver our AI platform as a service to support drug developers translate candidates with certainty. VeriSIM Life leverages the combined power of AI, knowledge-based Quantitative Systems Pharmacology (QSP), and state of the art computational chemistry models to enable predictions regarding the likelihood of translational success of drug candidates from just the chemical structure of the molecules and the knowledge of the target - without the need for any in vitro or in vivo data for the specific candidates. Using our hybrid AI models, we predict a Translational Index™ - a multimetric index that combines efficacy and safety aspects of drug candidates at different doses in humans. Translational Index ranks candidates in terms of their likelihood of clinical success as well as guides dose optimization. Think of it as a “credit score” for drug development, and the higher the number, the more promising the drug candidate.

The figure below shows an example of our platform predictions to guide candidate and dose optimizations. The candidate with ID of 2 shows the highest Translational Index™ score at a lower dosage. However, that candidate displays severe off-target toxicity aspects at a higher dose that one must consider. 

Reach out or book a demo of BIOiSIM today.