Scientists used to think computers were just additional tools for data processing — nowadays, the idea of fully virtual experiments is becoming popular. For now, the aspirations of medics and biologists focus only on creating a digital human twin to simplify drug design and make personalised therapy a reality.
In silico experiments: how do they work?
The idea of computer-based experiments originated in the 1990s — that was when the in silico term (literally “in silicon”) was coined. Now, it is widely used alongside terms like in vivo (“within the living”, i.e., experiments in a biological system) and in vitro (“in the glass”, i.e., test-tube studies in an artificial environment).
The concept of in silico experiments involves uploading input parameters to a computer to get modelling results that correctly describe the behaviour of an analysed system. The in silico approach now comprises several levels, including the modeling of specific molecules, biochemical processes, and physiological systems.
In particular, in silico methods encompass molecular modelling — a set of computing methods for calculating molecular systems of various complexity. Those methods can analyse amino acid or nucleotide sequences or study extensive databases (the bioinformatics scope of application).
Another method studies the evolution of molecular systems via empirical force fields, i.e., investigating the spatial structure of protein molecules based on structural templates of proteins with — homologous amino acid sequences. Other approaches analyse interactions like ligand-receptor and enzyme-substrate pairs; categorised as molecular docking, they are applied in drug design.
At present, in silico experiments cannot fully replace human models, but this direction seems highly promising. One of the reasons is that computer research can use virtual models, saving time and cutting costs on in vitro and in vivo tests.
For example, it is possible to design a molecule in silico to match a specific fragment of the target molecule, with which the drug needs to form bonds: this approach is called de novo drug design. Other methods predict the reaction of a substance based on its chemical structure, thus helping assess the toxicity of reactive impurities.
Almost all current in silico models use already available data on substances and their toxicity, collected from existing databases. For instance, PMI company (that actively applies computer models for their tobacco research) launched INTERVALS, an open online platform for joint work with big data that facilitates the planning of new experiments. Accessible information on previous studies and protocols helps scientists verify new hypotheses and conduct their research faster.
Digital twin startups
In recent years, clinical trials involving computer-modelled virtual patients have become more common, enabling scientists to study rare phenotypes that can be hard to select for real-life trials. Besides, in silico tests make it possible to compare alternative treatment methods on the same patient, reducing the timeline and the sample size.
For example, a research team from the UK and the Netherlands, led by Alejandro F. Frangi (the University of Leeds), proved that trials involving a virtual cohort replicate findings of conventional trials on real patients.
The scientists modelled the results of brain aneurysm treatment with a flow-diverting stent, using a sample of virtual anatomies selected from clinical databases. A stent is a flexible tube inserted in the artery to reduce the blood flow and lower the risks of rupture and ischemic stroke.
Eighty-two virtual patients were selected considering their age, sex, nationality, physiology, anatomy, and biochemical characteristics; the researchers needed to validate that all the parameters were similar to those of real participants of stent trials. Then, they developed a model that analyses the stent’s impact on blood flow in each “digital” patient.
The researchers compared the in silico experiment findings with the results of three traditional clinical trials. The virtual study showed that the treatment would be successful in 82.9% of cases, while the success rate of the conventional trials was 86.8%, 74.8%, and 76.8%.
Scientists at the University of Michigan Medical School developed HostSim, a virtual model of a Mycobacterium tuberculosis host, to analyse the immune response to the pathogen and predict the progression of the disease. The researchers had to use the virtual model because the data from sites of infection (lung granulomas) are generally unavailable for collection. The artificial model of a primate host consisted of lungs, LN, and blood: the physiological compartments making up the majority of dynamics that occur during disease progressions.
A few years ago, Dassault Systemes company (France) launched The Living Brain and The Living Heart projects on the 3DExperience platform. Based on the information provided by patients and collected during studies, the developers created a digital model to simulate blood circulation and test various treatment plans. The Living Heart is now applied for designing new medical gadgets, verifying drug safety, and preparing personalised surgical methods.
Meanwhile, The Living Brain Project is used for researching epilepsy and identifying the areas of the brain that cause seizures. In 2019, the FDA extended its cooperation on the project with Dassault for five years.
The Ebenbuild company (Germany) offers personalised digital twin therapy for patients with Acute Respiratory Distress Syndrome (ARDS). When developing a digital twin, the company specialists use individualised lung ventilation parameters to increase the patient’s chances of survival and recovery. All the data are extracted from CT scans via AI and image analysis methods, creating patient-specific lung segmentations of the lung.
“Local mechanical overload of the lungs due to suboptimal ventilator settings is a major contributor to the high mortality in patients suffering from Acute Respiratory Distress Syndrome (ARDS)”, Ebenbuild experts elaborate on the company’s website. “Our technology enables us to provide the best possible protective ventilation protocol for each patient, reducing ventilator-inflicted lung damage. Combining a CT scan of the patient’s lungs with in-depth physiological knowledge, engineering, and physics-based algorithms, we create highly accurate digital twins of the human lungs.”
Combining a CT scan of the patient’s lungs with in-depth physiological knowledge, engineering, and physics-based algorithms, we create highly accurate digital twins of the human lungs.Edenbuild – a company providing precision therapy based on personalised digital twins
Are human digital twins a possibility?
The human digital twin was ranked second in the LIFT Radar 2021 rating of the most popular digital healthcare technologies. In 2020, it was ranked third in the IEEE Computer Society rating. In 2020, Gartner, one of the world’s leading analytical companies, claimed that digital human models would significantly influence society, health, and business in the next decade.
We must admit that this method is only making its first steps in the medical industry, namely in drug development and personalised treatment. Meanwhile, virtual models are actively used in construction, manufacturing, automobile, and aerospace industries, helping design systems, manage operations and predict equipment wear in real time.
An interest in digital twin technology is also growing thanks to the evolution of various medical gadgets for data collection. The thing is that creating a virtual patient is impossible without regularly updated information on genomics, biomics, proteomics, and metabolomics, as well as physical markers, demographic, and lifestyle data over time of an individual.
In the best-case scenario, this vast array of miscellaneous data must be constantly “fed” to the virtual twin online, providing medical practitioners with accurate information on the patient’s health status and enabling them to find the most appropriate therapy.
In 2019, Chinese scientists Lui, Zhang, Zhou, and their colleagues described the prerequisites for digital twin software. First, digital twin models should be established with the help of advanced modeling techniques or tools (SysML, Modelica, SolidWorks, 3DMAX, or AutoCAD). Second, data connection should be done in real-time through health IoT and mobile internet technologies. Third, the simulation should be validated by regular calibrations. Finally, the results of virtual model manipulations (e.g., diagnostics) must be submitted to the patients.
If science learns to create virtual human twins, this breakthrough will undoubtedly usher in a medical revolution. Doctors will be able to introduce personalised treatments and disease prevention, applying diagnostic and therapeutic methods tailored to individual patients and their needs based on their genetic, biological, phenotypical, physical, and psychosocial characteristics.
Weak spots of in silico methods
At its present level of development, in silico methods are unable to analyse all impact factors of a new drug. That is why computer tests cannot fully replace conventional clinical trials and are currently used in drug design to reduce the number of animal models and make pre-trial research faster and less expensive. Besides, computer experiments have not gained absolute trust and must be verified by classical methods.
No matter how attractive the in silico may seem, many still doubt that processes in living cells can ever be described so precisely that scientists will be able to model life with their hands. For now, people oppose an overly mechanistic, “soulless” approach to the wonders of nature.
Even if we assume that the mystery of the human body can be solved, scientists will still need more sophisticated technologies for collecting and storing patients’ data. Also, medical gadgets should be more available so that the existing social and economic gap does not grow wider.
Researchers will also need to pay special attention to the quality of their data. Virtual patient models require objective data sets; however, many data arrays do not meet this requirement at the moment. For example, there is an evident racial and gender bias, as white men are represented in the studies more often than any other group.
Overall, data collection and storage remain extremely challenging tasks. Digital twins require detailed data sets and unified electronic health records to ensure proper automated collection and analysis of medical information. As of now, electronic health records are too heterogeneous, and their data are siloed and unstructured. Apart from that, it is also necessary to address the ethical matter of patients’ willing consent for various operations with their personal information, as well as the issues of data confidentiality and security.
Meanwhile, developers of digital twin software must ensure that their interface is user-friendly and facilitate communication between the medics, the patients, and the computer.
There is also a looming risk of eugenics. Virtual twins can demonstrate which genetic profiles have better survivability — and thus exhume the arguments about “good” and “bad” genes. As a result, there is a threat that people will start selecting embryos based on their profile, and employers will ask candidates about their gene pool at job interviews.
Finally, another pitfall is the general mistrust in AI-made decisions, demonstrated by patients and medics. Recent studies on AI integration into hospital environments have revealed that doctors are skeptical about AI algorithms, expressing concerns over potentially inaccurate diagnoses, wrong treatment, and a chance of machines eventually replacing human specialists.