30 years from now: what will clinical researches look like? Rustam Gilfanov’s opinion
An effort to look into the future and find out how quickly new drugs can be developed.
These days, it takes from 8 to 14 years to do preclinical and clinical trials before medicines appear on the shelves of drugstores for the first time. Besides, it costs at least 1.3-2 billion dollars to test a new compound. 80% of overall expenses are preclinical studies and clinical trials. Moreover, according to statistical evidence, only one out of five compounds reaching Phase I of trials will subsequently be approved.
How can this process be facilitated, and how can we save money? This article is an attempt to make a prediction of what clinical trials will be in 30 years.
The models crisis
It still takes a lot of time and money to develop a drug, and the probability of success is still low. According to scientists’ estimation, average investments in this area of research are 1.3 billion dollars per drug, while the approximate time of development ranges between 5.9-7.2 and 13.1 years for non-oncology and oncology medications, respectively.
Only five out of 5,000 compounds reaching preclinical studies make it to Phase I of clinical trials. The general approval rate of trialed drugs is not more than 13.8%. The main reasons for clinical trial failures are toxicity and low efficiency.
Those failures often happen because of poor choice of models: compounds that displayed great potential during animal tests happen to be inefficient when used in relation to people. This is the reason why one of the main trends of today’s drug development is to deeply analyze the level of the model adequacy when it is used for preclinical research (regardless of what it is: a rat or a macaque) as compared to the human body.
In spite of the fact that most failed drug candidates are inefficient but harmless, the selection of irrelevant models may sometimes cause disastrous effects. For example, the case with the TeGenero company. It was working on a new medicine against autoimmune disorders. TGN1412 (Theralizumab), a humanized monoclonal antibody, showed wonderful results during the preclinical trials when used on monkeys. Although, human volunteers who participated in Phase I trials suffered a lot from a severe cytokine storm an hour after administering the test drug. More than that, some participants even went into a coma.
The incident was caused by a simple reason: it happened that safe TGN1412 dosages for monkeys and people are entirely different. At the same time, this proves that animals can be only provisional prototypes of a human body.
As for the effectiveness of new medicines, drugs against cancers and CNS disorders take the longest time before being introduced to the market. According to statistics, they comprise half of all failed clinical trials. Two factors: the complexity of the diseases and the multiple points that have an impact on their progression make animal tests of new medicines almost meaningless.
3D models and organs-on-a-chip
One of the ways to fill in the gap between human beings and animals is to use three-dimensional models of human organs that efficiently recreate the states of a human body due to the environment of neighboring cells and the extracellular matrix.
For example, 3D models help to evaluate a more accurate dosage during cancer drug trials. External cells obtain a higher dose than those within the culture, like cells in a living body (especially in the case of solid tumors).
At the same time, 3D models cost a lot and are difficult to design and use. That is why, in spite of their functional benefits, they are not as popular among researchers as two-dimensional versions. However, they might become commonplace three decades from now.
Organs-on-a-chip (OOCs) is a more sophisticated model. The devices cultivating cell cultures, simulating mechanical and physiological responses of organs and even organ systems, can combine a 3D organ model with a microfluidic platform. Researchers have already made up the OOC models of microvessels, kidneys, lungs, liver, intestine, and their combinations.
How can such a model be created? For instance, preparing a 3D culture for modeling bronchi and smaller airway tissues demands primary bronchial human cells. The cells are cultivated in culture flasks and then seeded on platforms with porous membranes that make it possible to place them at the air-liquid interface and differentiate the original basal cells of the bronchial epithelium.
Actually, the concept of organs-on-a-chip occurred for the first time a decade ago. However, it did not become widespread because of its complexity and the strict legal regulations that were imposed on the pharmaceutical market.
At the same time, some companies help Big Pharma to reduce their costs and decrease the risks of unsuccessful trials by suggesting OOC solutions. For instance, the Emulate startup is a partner of several pharmaceutical companies that design chips studying immune responses to CNS disorders. The company also developed the liver-on-a-chip model for analyzing microbiome effects and carries out toxicological research when developing the drug.
The chips can deliver the data previously obtained through animal or human tests. This technology has the potential to revolutionize the industry and entirely refusing preclinical trials involving animals. In the best-case scenario, researchers will make a presentation of the “human-on-a-chip” model in 30 years and introduce individualized preclinical studies.
Omics data when designing a drug
Some points associated with preclinical trials can be resolved by omics data which are vast pools of molecules (they are collected on various levels of biological processes). These molecules demonstrate the state of a whole body or its part. Omics data are analyzed by bioinformatics which is an interdisciplinary area combining statistics, biology, and computer science.
Omics data are mainly applied at the earliest phases of preclinical trials to estimate the features of candidate molecules by using the information obtained from cells and to get a “mold” of a sick organism for testing the influence of various substances.
A classic situation of the same-level omics data use involves the comparison of tissue transcriptomes of healthy and diseased people. Researchers study the expressions of the gene for determining damaged molecular pathways. Afterwards, they choose the candidate molecules that can have an impact on the target molecules.
The latest trend involves the analysis of multi-omics data, i.e., the information on peptides, metabolites, proteins, and RNA and DNA sequences. This assists in seeing the entire picture of a disease, because pathological processes have a negative impact on multiple cell levels.
Ideally, it will become a standard practice 30 years from now to collect various data arrays at the start of preclinical studies to identify disease progression patterns, determine and test potential targets, choose and check candidates, and leave out toxic molecules. This is sure to make drug design much quicker and more individualized.
Moreover, it will make it possible to control the entire process (except for animal tests or cell modeling) via a computer. Provided the fact that expressions of some disorders have already been published, it will be necessary only to compare them with healthy tissue expression profiles and find out about drug properties. There are already databases that offer such information, e.g., ChEMB and L1000 Viewer. The latter allows scientists to learn what gene expressions have an impact on certain medicines. This algorithm has helped make a prediction of the effectiveness of six candidate drugs for treating gastric cancer.
Insilico Medicine, the world-famous international company, made a success of this approach by announcing its preclinical studies of a candidate molecule for kidney fibrosis treatment [100 that showed the required pharmacological properties both in vivo and in vitro]. The entire process took 18 months. At the same time, research similar to this generally needs years. Before that, the company had also used omics data to find a new target for idiopathic lung fibrosis.
To sum it up, omics data can bring us closer to individualized medicine. Probably, three decades from now, doctors will analyze their patients’ omics data to search for the right drug, observe their reaction to treatment, and change the prescription or dosage if it is necessary.
Machine learning and computer vision
Studies on disorders of the central nervous system comprise the second-highest number of failed preclinical trials. Since 2011, R&D departments of large companies have constantly been canceling their research even on diseases like Alzheimer’s.
The problem is that CNS disorders progress depending on environmental aspects and hereditary and may vary in severity and symptoms. More than that, preclinical trials do not even try to recreate all the influencing characteristics.
For instance, the main risk factor for strokes is age. At the same time, most in vitro research is done by applying cells of “healthy male mice.” When testing the effectiveness of new medicines on this dubious sample set, researchers ignore bad habits or comorbidities. Consequently, we should not forget the general biological differences between people and mice. The models used for tests are too distant from reality.
Moreover, the Diagnostic and Statistical Manual of Mental Disorders proves that, unlike animal counterparts, people have a lot of various behavioral patterns. During preclinical trials, these patterns are underestimated and are approximately assessed in terms of basic animalistic reactions. How many times did the most stop running? For how long did it stay near the wall of the cage? What is the maximum speed at which it can run in circles?
AI technologies can assist in studies like those to determine some verisimilitude. For instance, some preclinical trials already apply behavioral phenotyping.
This method uses the automated analysis of videos that display moving animals. Special programs define behavioral patterns that are based on body positions, typical movements, and gestures.
MoSeq (Motion Sequencing) is one such algorithm. It is used to test neuro- and psychoactive drugs. Watching hundreds of 20-minute videos, the algorithm can identify the pharmacological class and dosage of a substance administered to animals. The behavioral reactions defined by MoSeq helped the researchers to determine previously unknown similarities and differences between them, as well as revealed adverse effects of administered drugs.
The number of software based on computer vision and machines will most probably increase over the next three decades and boost the efficiency of preclinical trials.
Moreover, methods of AI and ML can greatly improve the success rate of drug development if they are applied to select clinical trial volunteers. They can also provide “smart” quality control of clinical data and on-site performance results.
Held back by the law
Legislation is still a critical factor that has an impact on the speed and use of technologies. Today’s approaches to clinical trials must follow detailed protocols to ensure compliance with scientific procedures and prevent potentially dangerous drugs from being released.
Even the most well-developed countries have laws that prohibit drug development based only on computer modeling. Animal tests and the participation of human volunteers are obligatory for any clinical trials, and AI methods can be used as supporting means only.
However, it seems clear the legislation will change soon when more and more success stories of AI-centered trials occur. By the middle of the 21st century, more than just several drugs that had in silico tests playing an essential role in their development will be introduced into the market. Even now, the latest technologies make the development process quicker and cost-efficient, and in the future, they will also make it more efficient.
To summarize, researchers keep using new ways to solve modern challenges and confirm the efficiency and safety of new medicines. They include both in vitro (3D cell cultures and OOCs) and in silico (computer models) methods. Maybe we will be witnesses of the times when the development of a new medicine does not make any living creature suffer, and its introduction to the market takes only months rather than years.
Rustam Gilfanov is a venture partner of the LongeVC fund, a venture investment fund formed to support startups in the field of biotechnology and longevity at the initial stages of development.