Mechanistic models serve as a useful candidate for modeling the detailed inner workings of a system in a true-to-life manner, going above and beyond simulating basic input and output. They can be used as a tool to improve preclinical translatability, reduce time and cost investments, and solve real-world problems safely.
At the most basic level, mechanistic models are simply replicas of real systems that outline the system’s behavior. They’re meticulously structured with a framework that demonstrates the interconnected network of the rendered biological system. The models are built based on applications of established biological, physical, and chemical laws.
Mechanistic models are incredibly precise and are structured to mirror equivalent variables in the real system they’re representing. The dynamics of the model’s behavior are also informed by the mechanism of the real system, with inputs and outputs related by mathematical equations. The models are useful for running experiments, making unique predictions, and closely inspecting the inner workings of a mechanism.
Mathematical models make complicated processes easier to understand. They can serve as a practical tool to turn complex biological systems into accessible representations. Once simplified, they can be used to form hypotheses, compare to experiments, and ultimately solve real-world problems.
As subsets of mathematical models, mechanistic models follow an analogous mathematical structure. As opposed to other kinds of mathematical models that rely on black box input-output, mechanistic models more fully outline the real system’s internal mechanism.
A mechanistic model could be used to gather data on the biomechanical reactions of a drug across various systems in the human body. Rather than just providing information about the input of the drug and the eventual biological output, this type of model would show the step-by-step details of how the output came to be.
In this same sense, mechanistic models are also related to biological models. As the human body contains thousands of proteins and genes, it'd be impossible to measure all of them.
When a mechanism of action is known, however, biological laws can be applied, such as the law of mass action. A model based on these known relationships can be established without needing to measure every individual biological component.1
In short, mechanistic models provide detailed information on internal processes. When looking at a blood pressure drug that decreases heart rate, a static model wouldn’t explain exactly how the heart rate was lowered. A mechanistic model, however, could explain how the drug entered the body, the specific receptors that it was bound to, resulting changes in hormone levels, and precisely how the hormones then modulated the heart rate.
A very clear example of a real-world mechanistic model is a flight simulator. Traditionally used by pilots in training, these simulators artificially reimagine the experience of flying a plane. The simulators are established based on mathematical equations.
Flight simulators rely on virtual reality and computer-generated images to reimagine the flight environment. They mimic the real systems of an aircraft, including the cockpit, displays and controls, the communication systems, and even conversations with others in the flight crew.
By using a flight simulator first, training a new pilot is safer, and has a greatly reduced financial burden. In this manufactured environment, novice pilots are able to safely make mistakes without involving real passengers, subsequently learning how to avoid them in the future.
Mechanistic models in the real world are also used for forming hypotheses, functioning well for improving the quality of first-in-human dose predictions. Such a model could utilize existing data on how a drug affects tumor size, then be used to simulate changing the dosage of the drug and witnessing the subsequent effect on the size of the tumor.
Mechanistic models can also be used to optimize and test automated decision-making systems. When working in theory, public safety, time constraints, and mounting costs are all in the abstract. In the real world, however, there isn’t always leeway for real-system experimentation.
If a company was testing a self-driving car that'd eventually be sent into rush hour traffic, for example, a mechanistic model could be used to simulate the applicable details of the driving environment, producing operative solutions that wouldn’t potentially injure other drivers, pedestrians, or passengers.
Just as a flight simulator cannot predict the weather on the day of a pilot’s first flight, all models have their limitations. As statistician George Box is famously credited for quoting, "all models are wrong, but some are useful."
Mechanistic models are established based on known biological mechanisms, but there are always unknown parameters that must be predicted. As is true in biological research at large, measurements and results have a level of uncertainty. Each experiment run will result in slightly variated output.
Another consideration is the computational expense of complex mechanistic models. The more detailed a model becomes, the more sophisticated a computing facility will need to be in order to properly run it. The computational power of the facility in choice should be taken into account when adding new caveats to an already complicated mechanistic model. However, advancements in scalable computational capacity (elastic cloud compute, GPUs, high performance software, etc), have put even complex modeling capabilities within reach for many organizations.
Bringing a drug from concept to market is a massive time commitment, after which a large percentage of compounds fail during the clinical stage. This is in part due to animal testing not always translating well to human application.
Combining mechanistic modeling with AI and machine learning offers an alternate solution, generating more precise data that improves preclinical translatability.
With their level of internal detail, mechanistic models can explicitly demonstrate a drug’s distribution from organ to organ throughout the human body. The models can be developed and expanded to include not only how a single organ reacts to a drug, but how each interconnected organ reacts in turn, from the liver to the brain to the heart.
Another benefit of mechanistic models in pharmacologic research is the ability to easily alter drug dosage and testing parameters from one experiment to the next. A wide range of specialized physiological parameters can be input, such as blood flow rates and organ volumes. Clearance mechanisms and drug solubility are also options for input.
With a mechanistic model in place, output data is accompanied by specific information on the internal biological processes of the simulated system. This reduces the need for experimenting with dosages in humans or animals.
Above all, mechanistic models answer the “how” question. They make studying the detailed inner workings of a mechanism both feasible and specific. With a mechanistic model in place, it’s possible to generate multiple rounds of detailed data, make informed preclinical predictions, and solve real-world problems.
What happens when you fuse mechanistic modeling with artificial intelligence? Read this case study.