Genome editing and artificial intelligence (AI) could revolutionize medicine globally. But to ensure responsible progress of genomics and AI in combination, a balance must be struck between promoting innovation and responding to ethical, social, and moral quandaries.
The term ‘genome editing’ has reemerged as a hot topic in the last five years. Recent breakthroughs in the gene editing technology, the clustered regularly interspaced short palindromic repeats, or CRISPR, have invigorated the biotechnology community with the promise of precisely manipulating the genome of any organism, including humans.
The new method found in CRISPR offers significant advantages over existing methods like zinc finger nuclease (ZFN) and transcription activator-like effector nucleases (TALENs). Experts now tout CRISPR as more efficient, simpler to use, and more able to edit many genes at once than older methods.
CRISPR edits DNA by acting as a pair of ‘molecular scissors’ to ‘cut’ and ‘paste’ DNA in the genome. To accomplish this, the CRISPR-associated (Cas) protein uses a short segment of RNA to find the genomics sequence of interest.
This feature makes this technology easy to use, as experts only need to change the guide RNA to locate different genes and can utilize the same CRISPR-Cas unit. Once the CRISPR-Cas system arrives at the target gene, the Cas unit can edit the DNA using a variety of mechanisms to insert, delete, or replace DNA at that site.
This function should prevent CRISPR-Cas from making edits on other genes, as it should only edit once it arrives at the target. The variety of genome editing functions this technology can perform also contribute to its wide applicability.
Some of the most significant applications of genome editing include medical uses for humans, as CRISPR could be deployed to repair disease-causing mutations. Practitioners could deploy this technology in the clinic in two broad ways, somatically or in germline modifications. Somatic therapy refers to editing the DNA in the cells of a human after birth, rather than during prenatal development
New types of therapy for existing diseases like cancer could appear when utilizing this technology, including editing immune cells to better target cancer. This application of CRISPR represents another kind of gene therapy, which is not unprecedented and presents less controversy than other uses of the technology.
Rather than simply treating cancer as it arises in adults, CRISPR promises the ability to edit out mutations from genes such as BRCA1 to lower the cancer risk of children even before birth. Genome editing similarly could correct mutations which cause debilitating genetic conditions including Huntington’s disease or cystic fibrosis, ensuring future generations will not suffer from such ailments.
However, techniques for germline editing pose safety and ethical issues both for the child born and for any children they have, as these genetic modifications will appear in subsequent generations.
Off-target effects, or the possibility of altering unintended parts of the genome, have presented an increased concern since a publication in Nature Methods suggested such effects occur more often than previously thought. Germline CRISPR use can result in mosaicism, where successful genome edits occur in only a fraction of an embryo’s cells.
Such non-uniform edits may fail to prevent a disease from occurring or could pose novel health complications. A recent development in methods involving earlier exposure of an embryo to CRISPR may aid in reducing off-target effects and mosaicism, though these groundbreaking findings will benefit from further study. Epigenetic effects may further complicate germline editing, as epigenetic factors augment an organism’s genome after conception to impact whether, where, and how the body expresses genes.
Somatic therapies may have effectiveness problems as well, as the human immune system could potentially (and correctly) recognize the CRISPR-Cas9 enzyme as non-human and develop antibodies that inhibit these gene editing tools.
The advent of molecular biology, genomics, and now CRISPR have promoted a conceptualization of biology as a machine. This view describes cellular and genomic functions using comparisons to engineering or software concepts. Scientists describe mitochondria as ‘powerhouses’, kinesins as ‘motors’ that move ‘cargo’, and insulin-producing microorganisms as ‘factories’.
Viewing biology as a machine arises from the core philosophy of synthetic biology – the engineering concept of design applied to assembling new biological systems to perform a task. Practitioners conceptualize building new organisms by using living ‘building blocks’ and genetically ‘programming’ the living machine to perform a wide spectrum of tasks.
This programmability of genomes in living machines allows for synthetic biologists to design novel functions into their biological machines, which may not appear in nature, and gives way to a problem-solving mentality towards living systems. Scientists have similarly embarked on work to utilize DNA for data storage, with a group succeeding in storing a movie and computer operating system in DNA.
This exposition of biology into computer science reflects a larger shift of conceptualizing biology as a machine, which can be programmed and designed to perform tasks, as any other machine.
Genome editing may have human medical applications, which could open possibilities in the ‘optimization’ of human biology. Viewing biology as a machine could result in somatic CRISPR therapies conceptualized as the equivalent of software ‘patches’ to improve an existing system. But CRISPR also may enable germline editing on human embryos, potentially yielding experts greater ability to design and optimize the ‘software’ of human biology at an early stage. A university hospital recently described genetically modifying immune cells to better combat cancer as “manufacturing” the cells and reported work on “the problem of manufacturing T-cells in a process that is easy to control, understand, and scale up.”
The presence of a mechanical conceptualization of biology amongst citizens and medical practitioners could promote the acceptance of CRISPR applications in humans. Viewing genome editing applications in therapeutics as software ‘patches’ or prelaunch ‘debugging’ could make this technology appear less foreign to the end-users.
Artificial intelligence (AI) represents an emerging technology with the potential to significantly impact the medical field. Progress in deep learning has enabled this shift, in which AI can learn by experience. This type of learning bears similarities to how children acquire new information, and the technology functions by utilizing program architecture that resembles the human brain.
Such an approach involves machine intelligence recognizing patterns in data and learning from its mistakes to better identify or classify new information. Deep learning specifically supports this function of AI by allowing the machine to recognize many layers of patterns; like identifying an animal by first recognizing its outline, then focusing on more specific details like fur.
This technology continues to find use in a variety of settings, from recognizing cats in YouTube videos to predicting RNA splicing patterns in mouse cells. AI technology can additionally utilize optimization tools and even contribute to new optimization methods. These capacities of deep learning to identify patterns from large datasets and optimize systems may provide a powerful tool when combined with genome editing
Due to advances in AI, this technology holds new potential uses in the clinic and in medical research. Deep learning software has demonstrated its ability to diagnose images of skin cancer and microscopy images, and predict disease in hospitals after reviewing patient medical records.
These AI learn to diagnose medical conditions by utilizing experiential knowledge, resembling how physicians are trained during residency.
Beyond the clinic, the technology can provide powerful analyses of currently available scientific information to model the molecular mechanics of how genetic variation causes disease.
Similarly, AI appears set to analyze a large number of genetic datasets. This capacity of deep learning to diagnose, predict, and find new patterns in how genetic disease operates could make AI an invaluable companion to clinical medicine. Access to cloud computing stands to further bolster these abilities as well as patient accessibility.
As AI continues to expand into the medical realm, the tool could have a powerful effect on how practitioners utilize genome editing. Since the new biology aims to optimize living machines to best perform their intended functions, gene editing practitioners may take this mentality to maximizing a patient’s ability to live a healthy life. Particularly if human biology is viewed as a machine, deep learning could find use in instructing experts on where to use technologies like CRISPR in the genome to ‘optimize’ the human body.
Limitations in AI applying deep learning to genomics and gene editing does come with limitations. While AI promise powerful new analytical and diagnostic methods in medicine, current machine learning software and their developers lack the capacity to explain how the programs arrive at their conclusions.
This curious ‘black box’ character of the technology could complicate risk assessments, as an inability to understand how AI ‘thinks’ creates issues in identifying possible areas where the machine could fail or make mistakes.
Should a program provide an incorrect diagnosis that influences patient and physician medical decision making, understanding why the technology made an error could pose a challenge. Furthermore, the quantity and quality of scientific information available could restrict the accuracy of deep learning projects attempting to model human biology and make predictions from those computations.
Additional accuracy challenges to these AI-based computational biology projects may also arise from an incomplete understanding of the natural plasticity of biology, obscured by such factors as epigenetics, gene environment interactions, and other variations between patients.
These informational bounds on the models produced by AI create a need for time-intensive research to confirm their conclusions about the clinical significance of genes, as the type and amount of information fed to deep learning software limits its potential outputs.
Validating AI predicted medical phenomena could rise in prevalence and importance moving forward, especially as the current knowledge of the clinical significance of disease causing mutations continues to evolve.
The concept of deploying AI and genome editing in combination to ‘optimize’ human biology could occur in the foreseeable future. Advances in deep learning already enable AI to contribute to the genetic diagnosis of cancer, through liquid biopsies, and provide actionable medical recommendations to patients using their medical history. And some genomics companies have already begun to pursue machine learning assets to boost their disease risk assessment potentials.
These developments suggest that AI could actively participate in medical decision making around germline genome editing in the future, especially considering the extent of data analytics required to interpret the human genome. Employing medical germline editing may occur in the relatively near future as well.
Applying gene editing to prevent genetic disease in this fashion could find significant benefit from AI tools which advise practitioners on where to use CRISPR in the genome to optimize health. The analytical power of deep learning could enable this technique by evaluating human biology and the genome similarly to an ordinary machine requiring streamlining.
AI optimization represents a value-free process, however, and would not independently incorporate moral or ethical judgments. This could exacerbate tensions over deploying CRISPR for human enhancement – the improvement of biological abilities above the level of the average person.
While public engagement prior to utilizing CRISPR to increase human biological abilities is called for, but defining enhancement and distinguishing it from using genome editing to optimize human health poses challenges. For example, increasing muscle mass above normal levels likely represents such an augmentation. But altering the human genome to unnaturally decrease a patient’s risk of cardiac disease presents a more challenging case to classify, as could many other uses of CRISPR to lower risk and promote health.
Such risk-lowering alleles would contribute to patient well-being and AI tools designed to aid in maximizing human health would likely include these genes as advantageous targets for CRISPR. Since current deep learning technology operates by identifying patterns rather than adhering to preprogramming, the software would likely be unable to appreciate the ethical nuances of enhancement versus disease prevention.
Viewing biology as a machine could further distort the line between deploying CRISPR for disease prevention versus for augmentation. Should AI determine that editing the human genome for increased health requires the use of risk-lowering alleles, the biology-as-a-machine perspective may express agnosticism towards applicable concerns over enhancement. Since a primary goal of this engineering mentality is to optimize its target system, conducting these genomic edits may represent the best way to achieve this goal and promote patient health. Arguments for germline editing may find use in justifying these actions, as edits which maximize health may represent a durable way to prevent disease and increases autonomy in the resulting child.
This perspective may also apply to CRISPR use resulting in more obvious enhancements, if these edits similarly produce optimized health. If deep learning analysis determines that genetical modifications which enhance eyesight also lowers the risk of ocular disease or if bolstering human brain function leads to later onset of cognitive decline, performing such modifications would both enhance the resulting child and optimize their biology for health. Such AI-enabled uses of CRISPR in the germline could challenge the significance of distinguishing between augmentation and disease prevention.
Just as a machine can be optimized to perform varying functions, gene editing in combination with artificial intelligence could enable the honing of human biology for purposes other than solely improved health. Recent genetic diagnostics which claim to predict the physical characteristics of a newborn could herald a time when this type of information becomes applicable to CRISPR germline editing. Though experts question the clinical validity of information about the genetic nature of physical traits or personality.
Combined applications of AI and CRISPR may be presented as a potential to optimize human biology for physical characteristics and not solely health, prompting more classical concerns for enhancement.
Relatedly, conversations about optimizing humans for space travel have already begun, including thoughts on reducing stature and boosting the body’s ability to respond to radiation. Genomic testing already exists to determine if an individual possesses genetic variants which would be valuable in space, a task which could be further augmented by AI analysis to identify variants which would maximize these attributes. Viewing biology as a machine influence these conversations as well, contributing to proposals to radically modify human biology to confer photosynthetic properties – a potentially useful trait for interstellar living.
Engineers can grant a machine new functionalities to better perform its tasks, a mentality which manifests in discussing adding photosynthetic capabilities to humans. Such biological additions would require a great amount of data analysis and optimization to prevent negative health consequences, and fortunately both would likely be assisted by AI tools.
These technological applications at the convergence of CRISPR and deep learning would require further strides in research and development, as current tools could not perform these functions. The connections between many genes and disease remain poorly characterized, especially given the countless alleles with unknown impact on disease which experts continue to discover.
The scientific community has thus criticized researchers claiming to have found relatively simple connections between genetic factors and complex diseases. Scientists have similarly disputed the merit of using genomics to give potentially overconfident counsel on preventing disease and improving health.
This evolving subsection of the genomics industry often offers genetic sequencing directly-to-consumers and can provide advice on anything from diet modification to boosting soccer performance.
Genetic tests which predict physical characteristics of newborns have received even harsher criticism from experts. And recent research in CRISPR suggests that adding a new gene to a human embryo may present more challenges than originally thought.
In experiments that tried to replace paternal DNA with a new gene, the embryo instead copied the DNA from the maternal gene, which could pose new challenges for editing embryos with two copies of diseased genes or for making progress in enhancement.
The ‘black box’ and insufficient data (quantity or quality) limitations on AI similarly restrict the possibility of machine learning optimization of human biology without further work. Advances in deep learning techniques alone may prove insufficient to improve the capacity and address the limitations of AI, and new approaches to the technology demand time and resources – potentially delaying the use of machine intelligence in genomics further. Ultimately, a successful medical product using germline editing (with or without AI) would almost certainly require more comprehensive genomics and machine learning research, substantial clinical testing, and the mitigation of remaining uncertainties.
While the United States seems to have been setting the tone in human genome editing, many other countries in Europe and Asia conduct work on CRISPR, with China as a notable example. Largely as a result of its industry’s work in DNA sequencing, China has emerged as an international giant in biomedical research and precision medicine in recent years. The Chinese biomedical science community may soon overtake US scientists in the number of research articles published in well-known academic journals, and China’s biotechnology industry has discovered an increasing number of drugs in recent years.
China has rapidly expanded its funding of synthetic biology research and could surpass American funding by 2020. Chinese scientists have aggressively moved forward with research on CRISPR in this context, and upset the international community in the last two years when groups reported using CRISPR in human embryos on three separate occasions. Suboptimal data transparency has further confounded efforts for outside parties to evaluate Chinese medical research on CRISPR.
The United States and China hold different social perceptions of the use of CRISPR in human embryos. The Chinese public largely responded positively to the use of genome editing in human embryonic stem cells by research teams in China. Both professional and social media celebrated the work for its contributions to the scientific field and most, but not all, did not express significant ethical concerns over the use of CRISPR. In contrast, many Western stakeholders maintain that these experiments demonstrate lax regulation of emerging biotechnologies in China. Others have argued China remained within appropriate ethical bounds while performing these experiments and note that the group received ethical approval from its home institution.
Current Chinese patent law prevents patents on human germline modification due to moral principles; however, other legal and regulatory structures in China may not be adequately prepared to address moral and ethical concerns arising from human CRISPR use.
These trends in biomedicine mirror the significant recent increases in Chinese investment in AI research, both at home and in international groups. Top technology firms in China employ machine intelligence technology of comparable caliber to leading American companies. Further, the Chinese computing industry has surpassed US efforts with two different supercomputers, hinting at a considerable potential for machine intelligence.
The prioritization of AI research may find use in the health care sector, as a new collaboration in China has embarked on utilizing this technology to interpret massive amounts of health information, including personal genomics. This resembles recommendations made by the Chinese Academy of Sciences to prioritize genomics research, and even to accelerate work in this field by exploiting computer modeling to optimize synthetic living systems.
Such policy recommendations for optimization of biological machines in combination with projects to use AI to understand genomic and other health data could foreshadow a time where AI-based optimization of human embryonic CRISPR procedures becomes acceptable in China.
Applications of genome editing with an AI companion may find uses beyond maximizing health. A genetics firm recently announced a newborn DNA sequencing service in China which will offer interpretations not only for disease risk, but also for various physical characteristics. This interpretation will include deterministic predictions about likely physical appearance, personality traits, and even more obscure items such as musical ability – following trends in emerging pseudoscientific genetic counseling.
Independent of whether this type of interpretation is useful today, providing and accepting genomic empowered predictions of physical characteristics could lead to considerations of gene editing at those loci in the future. Social acceptance of such medical predictions or interventions in China could press Chinese AI-based CRISPR use to optimize human biology for physical characteristics, not solely for improved health. The possibilities for this type of optimization could have further ethical consequences should scientists better understand the genetics behind cognitive function and emotional capacity, with recent studies finding some genetic connections to intelligence and empathy.
Should China permit the use of CRISPR germline editing, with or without AI analysis, citizens of other nations could potentially access this technology through medical tourism. The concept of medical tourism involves patients traveling to another jurisdiction with the intent to receive medical treatment, and could have a role in the accessibility of germline genome editing. High-profile medical tourism for embryonic procedures has already occurred in recent years with mitochondrial donation, a technique which functionally replaces diseased mitochondria in a human embryo with healthy ones.
In 2015, an American physician performed the technique for a Jordanian couple in a Mexican medical institution. Mitochondrial donation was not legally permitted in many Western nations at the time, though no prohibitive legal barriers existed in other jurisdictions like Mexico. Reports exist of mitochondrial replacement occurring in other nations, including China and Ukraine.
The reality of patients and physicians engaging in medical tourism to perform embryonic treatments with emerging biotechnologies suggests the same could occur with CRISPR germline editing. Obtaining access to a medical intervention that is illegal in one’s jurisdiction and not another can motivate patients to participate in medical tourism.
