Personalized Medicine: Tailoring Healthcare to You

Personalized medicine is a groundbreaking field that can transform how we approach disease prevention, diagnosis, and treatment. It’s also known as precision medicine and individualized medicine. It involves making healthcare decisions based on the unique genetic and genomic data of each individual patient. Personalized medicine recognizes that each person’s genetic composition, lifestyle, and environmental factors all contribute to their health outcomes, as opposed to a one-size-fits-all approach.

The concept of personalizing medicine rose in the 1990’s, assisted by the Human Genome Project (HGP), which decoded over three billion base pairs of the human genome, which provided researchers around the world essential genetic information. The International HapMap Project added further fuel by identifying genetic variations linked to some human diseases. These two efforts helped to better explain things that were known but little understood, such as why some drugs worked better for some patients than others, and why some patients experienced severe side effects. Medical researchers were now able to understand the molecular factors that are influenced by individual genetic constitution, and we gained two new fields of medical research: pharmacogenetics and pharmacogenomics. These fields study genetic causes behind drug responses and the impact of genome variations on the treatment outcomes.

Before I get into the medicine part, I think I need to explain the difference between genetic data and genomic data. Genetic data focuses on specific individual genes and how the affect our traits and conditions that are handed down through generations. Think green eyes and blonde hair that you got from your dad. Genetic data is like looking at individual puzzle pieces. Genomic data takes in all of a person’s genes, and it considers how the genes interact with each other, and with their environment. It’s like the whole puzzle, but not just the puzzle; it’s also the table it sits on, and the room that the table is in. Humans are 99.9% genetically identical, but that last .1% is where the most vital clues lie regarding the causes of disease and personalized health.

I’ve written several pieces on Artificial Intelligence (AI) and the changes it’s making in our world, but AI really changes the game in personalized medicine. AI can process huge data sets, including genetic information, medical histories, and lifestyle factors, as one group of data, incorporating it all into a coherent picture. By analyzing the disparate data, AI can help create personalized treatment plans for patients. Additionally, going beyond merely treating symptoms, precision health aims to address the root causes of diseases, taking into account a patient’s genetics, lifestyle, and environment. The convergence of AI and personalized medicine represents a paradigm shift toward more precise, predictive, and preventive healthcare. As we stand on the cusp of this new era, the potential for improving patient outcomes and healthcare efficiency has never been greater.

The Evolution of Personalized Medicine

Personalized medicine has evolved from a one-size-fits-all approach to targeted therapies, powered by several breakthrough events. We can go back as far as 1901, when University of Vienna researcher Karl Landsteiner identified what we call the ABO blood group system (“What’s your blood type?”). Besides helping the medical field come to understand why some blood transfusions succeeded and others failed, it led the way in recognizing differences in the biology of each patient. The following year, Sir Archibald Garrold was able to make a connection between Mendel’s laws and genetic inheritance of a particular disease inherited only when both parents carry a certain recessive gene.

DNA discovery work was going on even before then, though. A chemist in Switzerland, Friedrich Meischer, analyzed used surgical bandages and came to understand the unique molecular structure each patient’s output carried. In 1919 in Russia, a biochemist named Phoebus Levene was studying the fit of DNA and RNA molecules with one another, and in 1953, the DNA double-helix structure was discovered by James D. Watson and Francis H. C. Crick. Finally, in 1977, Frederick Sanger revolutionized the ability to explore genetic information when he developed the first DNA sequencing method.

After that, personalized medicine entered the modern era. In 2003, the completion of the HGP project to decode the entire human genome gave us more essential genetic insight, and it expedited the evolution of personalized medicine. We are able to interrogate the whole genome and obtain panomics data – all of the puzzle pieces that make up you: genes, proteins, metabolites, and all of the tiny organisms that live in your body. The results of all of these advances and discoveries brings us to the point of being able to shift from a cookie-cutter approach to much more proactive care. We are now able to put the focus on prevention and particular drug therapies that can adapt to each patient’s unique medical attributes. One benefit is a reduction in reliance on trial-and-error in prescribing medicines.

Today, we can see personalized medicine in many examples. Doctors can analyze certain tumor markers in a patient’s blood or tissue and develop a tailored cancer treatment. The markers can help identify the type of cancer, its stage, and potential options for treatment. Doctors can also read a patient’s entire DNA sequence, which can help predict risk of certain diseases, identify some genetic variants of specific conditions, and recommend preventive measures. Using the same DNA sequencing, a patient’s potential reactivity to certain drugs can be predicted, allowing doctors to rule out those that may be dangerous or ineffective for that patient.

The Impact of AI on Personalized Medicine

I’ve written earlier about how we can use AI to help process vast amounts of information in general, and that certainly applies to using it to personalize healthcare. Now I’m going to give you some specific examples. These aren’t the only possibilities, and we should expect even greater expansions of the application.

Genetic algorithms can imitate the process of natural selection that Charles Darwin observed and documented in the Galapagos Islands. They start with a group of potential solutions to a problem. The algorithms iterate through the possible solutions and scores each on how well it solves the problem. A predetermined number of higher scores will be selected for advancement. Genetic algorithms evolve the population or group toward optimal solutions over time, by mixing parts of two solutions (crossover) and introducing random changes (mutations).

Neural networks are like interconnected virtual brains. Deep learning algorithms use neural networks to find patterns in genetic data, which would be impossible without the algorithms. The neural networks learn to identify and pull out relevant features from the raw genetic data sequences, which helps them find important genes or mutations. Through repeated cycles, the algorithm “learns,” and, over time, they are able to predict risk of disease, responses to different medicines, or personalized treatment options.

Finally, there’s a process called Clustering and Dimensionality Reduction. K-Means Clustering is organizing a big group of objects into smaller, related or similar groups – like sorting your clothes by fabric type before washing them. Next, the algorithm picks through the groups of data looking for relationships between data points. This is kind of like working a jigsaw puzzle using a methodology. I start with the edge pieces first (and if you don’t do it this way, you’re borderline psychotic). Once I have those, I look at the image of the finished puzzle for elements that will be easy to identify – a house with distinguishable lines, a brick walkway, bold colors. At some point, you don’t need the picture anymore, and you know what the picture is. In an episode of The Big Bang Theory, the group is on a scavenger hunt for clues to a prize. The first clue is a jigsaw puzzle. Penny and Sheldon are a team, and with less than half the puzzle completed, Penny identifies the picture as the comic book store. It’s like that. By clustering and reducing the dimensions, researchers can gain valuable insights into genetic variations and subtypes of diseases.

I’ll give you a couple of examples of AI identifying patterns and making predictions. Some programs, I’m sure you’ve heard, can recognize faces, in varying degrees of accuracy according to a number of factors. However, there are also programs that can identify genetic disorders. Certain facial characteristics, for example, may indicate Down syndrome or other genetic syndromes. Other machine learning techniques can analyze samples of bodily fluids in order to predict the primary type of cancer a patient may have. By understanding the cancer type and progression, doctors are guided in the treatment plans for these patients.

AI can play a strong part in the discovery of new drugs, as well. Deep learning models using neural networks are able to analyze vast amounts of chemical data. Then they can predict how potential drug compounds will behave biologically and optimize their properties. Bringing a new drug to market represents a huge investment in time and money, but using AI can reduce the needs in both. AI can also simulate interactions between drugs under development (the compounds being considered for inclusion in the drug) and the biological targets like proteins or nucleic acids. The AI can select the most appropriate combinations rather than having to physically test thousands of compounds. Additionally, rather than estimating drug dosages and combination, AI can personalize them for individual patients.

Now we come to the clinical trials for drugs, which is where the drug developer has patients actually using the drugs. Before dispensing the drugs to the patients in the test, AI can help design the study so that the study provides relevant information to the researchers. AI can do the work of defining and refining criteria for eligibility, and then it can identify the relevant patient groups and subgroups. This can accelerate the study dramatically over having this work done by humans. It also can help reduce the burden on the patients in the study by designing the test with fewer visits to the test center, and it can streamline the application and acceptance process for potential test subjects. Having a well-designed plan and only the best-eligible patients can also minimize the changes that researchers may have to make after the study has begun. These aspects all contribute to a more efficient, and more important, a more successful and accurate study.

Challenges and Ethical Considerations

There are several ethical concerns regarding AI in general, but when it comes to healthcare and privacy, the stakes rise dramatically. None of us wants just anyone to be able to have access to the most personal information that exists about us, and there’s nothing more personal than our unique genetic information. It’s not just data about who we are, but who we were (our ancestry) and who we may become (disease risks). Unauthorized access to or improper handling of this information can open the doors to misuse, stigmatization, or discrimination. However, the massive benefits that researchers can achieve from access to genetic information must be considered in the ethical conversations. It’s always going to be a challenge to balance the two vital needs of privacy and access to information.

It’s also important to understand that AI “learns” from data it receives, some of which may not be tilted in a particular direction due to existing bias in current systems. For example, if a survey asked respondents to determine the color of a sample of paint, and the only options were “red” and “blue,” the AI model would learn that the color of the pain sample is whatever more of the respondents determined it is – even if the color is actually purple or green. It’s not infallible, and it’s only as good as its training. Our current healthcare system contains factors that do not allow equal access to care, and the experience across demographic segments is very different, which can “mistrain” the AI model. Feeding biased information into an AI model will perpetuate and accentuate that bias in the resultant output. It will always be a difficult task to ensure that the input data is diverse and representative to the work. To address the potential bias issues, researchers can use fairness-aware algorithms that are able to consider such protected attributes as race and gender during the AI training process. Being able to explain how the AI arrived at its conclusions can help reveal any existent biases. Additionally, researchers will need to perform regular audits of the AI models they use to find and correct any disparities in outcomes.

Finally, using AI to develop personalized healthcare is going to have to pass rigorous regulatory approval processes. The U.S. Food and Drug Administration (FDA) will have the authority and responsibility to ensure that products and processes adhere to safety, efficacy, and quality standards. While the AI can drive innovation and advancements, it will be regulatory compliance that assures the public of patient safety. Striking the right balance between pushing the boundaries of AI capabilities and meeting regulatory requirements will be crucial for successful adoption in healthcare.

Case Studies

One groundbreaking area where AI has been successfully integrated into personalized medicine is in the research and treatment of cancer. AI algorithms can analyze a patient’s genetic data, images of tumors, and the patient’s medical records to develop cancer treatments specifically for each patient. For example, they can more accurately predict which therapies will be most effective based on individual patient genetic profiles. This means that the patient will receive therapies tailored to him/her, which can minimize unpleasant side effects while improving the outcome.

In another example, by examining patient genomic data, lifestyle, and medical history, AI models have become adept at identifying early warning signs of Alzheimer’s and heart disease. By seeing these warnings early, caretakers and healthcare providers are able to deliver more timely interventions. Possible preventive measures for some conditions can significantly reduce healthcare costs. Ongoing research and experience is demonstrating the importance of interdisciplinary collaboration, and it also highlights the criticality of data quality and ethical application.

The Future of AI in Personalized Medicine

On the horizon of AI in personalized medicine, we can already see research in biomarker testing for information on certain diseases like cancer. In conjunction with biomarker testing, we’re also going to see advancements in targeted therapies based on individual biomarker profiles. Research is also ongoing with nanorobots, tiny robots that can navigate about the human body at the cellular level. The hope is to be able to deliver drugs precisely to tumor sites, being able to repair tissue damage without surgery, and monitoring health more intricately than we can now. Advancements in our current range of wearables, like smartwatches and fitness trackers, will allow the collection of more detailed biological signals. Future developments may include medical alert notification and interventional delivery of certain medicines.

To realize the maximum potential of the possibilities, though, there are some challenges that must be met. One is that AI models rely on high-quality, diverse data in order to make accurate predictions. Datasets that contain biases or that are incomplete will provide skewed results and further perpetuate the disparities that exist in our current healthcare systems. Another challenge is the ongoing tug-of-war between privacy and public interest, but that’s not a challenge that’s limited to this discussion. It is a concern at every level where personal data is involved, but when we’re talking about the very genetic code of what makes us who we are, have been, and may become, it becomes critical to address. It’s a discussion that cannot be postponed, but must occur at the beginning of any endeavor where it comes into play at all. Finally, there’s a gap that needs to be bridged between medical expertise and AI technology. It can be hard to teach these new methods when the teachers were all trained on more traditional methods, and AI carries a scent of untrustworthiness due to early errors. Both the errors and the prejudice against AI will need to be overcome for the potential to be realized.

Wrapping it Up

That was a long read, wasn’t it? I hope I broke a difficult subject down and made it enjoyable to consume. I’m fascinated by the prospect of early identification and treatment of conditions that have historically been impossible, difficult, or unpleasant to treat. I’d love to hear from you: What are your hopes, expectation, or concerns around using Artificial Intelligence to personalize healthcare?

I’ve compiled a list of sources I used to pull this together, that you may find interesting for further reading on the subject.

https://binariks.com/blog/ai-machine-learning-for-early-disease-detection/

https://blog.algorithmexamples.com/genetic-algorithm/applying-genetic-algorithms-in-ai-a-how-to-guide/

https://bmcbioinformatics.biomedcentral.com/articles/10.1186/1471-2105-13-24

http://genetics.bwh.harvard.edu/courses/Biophysics205/Papers/Primers/PCA_primer.pdf

https://journals.plos.org/digitalhealth/article?id=10.1371/journal.pdig.0000022

https://medriva.com/news/medical-breakthroughs/the-growing-trend-of-fda-approved-ai-technologies-in-medical-practice-what-you-need-to-know/

https://medium.com/aimonks/the-future-of-personalized-medicine-ai-precision-health-3a0291a1a394

https://news.umich.edu/health-care-artificial-intelligence-gets-biased-data-creating-unequal-care/

https://sitn.hms.harvard.edu/flash/2018/understanding-ownership-privacy-genetic-data/

https://stepofweb.com/ai-drug-discovery-innovation/

https://themillsinstitute.com/personalized-medicine-examples/