Personalized medicine, also known as precision medicine, is an innovative healthcare approach that seeks to customize medical treatment to an individual’s unique characteristics. This emerging medical model considers differences in patients’ genes, environments, and lifestyles in designing tailored prevention, diagnosis, and treatment approaches. Alterations in the genetic makeup can disrupt biological pathways leading to diseases, and genetic information provides a new level of understanding of the intricate networks of molecular mechanisms involved in disease biology [1]. As advances in genomics technologies exponentially decrease the costs and increase the throughput and quality of genomic data generation, the application of such technologies to population healthcare is envisioned to lead to major improvements in healthcare outcomes. This includes the development of better diagnostics, drugs, and drug combinations for particular disease subtypes and patients and a shift from reactive treatment based on population norms to proactive and preventive initiatives based on individual risk.
There is a growing recognition that integrating multi-dimensional data and knowledge streams generated by emerging medical technologies encompassing omics, imaging, and electronic health records (EHR) can provide a deeper understanding of individuals and populations pivotal to personalized medicine and improved population health. However, a complex interplay of challenges exists that needs to be effectively harnessed to realize a vision of better care and improved safety and overall health outcomes at lower costs, including empowering individuals to participate in their health management. Advances in pervasiveness and affordability of mobile personal devices along with interconnected sensing technologies, including genome/single nucleotide polymorphism – SNP arrays, lab-on-a-chip devices, multi-mode (e.g., wearable, implantable) biosensors, and P4 modeling can assess the multi-modal phenotype of individuals, including genome, microbiome, proteome, metabolome, and behavioral information as well as environment measurements. Such information needs to be augmented by interoperable and integrated clinical record systems that can track EHR across time, location, and practices. On the one hand, there are concerns and fears regarding individual privacy, ownership, and commercialization of data. On the other hand, a criticism to big data initiatives is mostly focused on computational complexity and preferences for data-driven analyses in unraveling biological knowledge and deciding health management strategies.
1. Definition and Importance
Personalized medicine refers to the practice of customizing healthcare decisions and medical treatments to each individual patient. This tailoring of treatment is based on a patients’ expected response or risk of disease. The goal is to bring better treatment efficacy and safety by shifting from one-size-fits-all and population-based testing to a more precise and targeted approach with increasingly broad applications [2].
The understanding of a patient’s biology, genetics, and environment is at the heart of personalized medicine. Using molecular diagnostics on a tumor, for example, can influence first-line therapeutic decisions and enable subsequent treatment adjustments. Gains in knowledge of biology and the rapid development of technologies, e.g., whole-genome sequencing, gene expression profiling, and proteomics, hold promise for expanding the birth of this emerging model of clinical care [1].
2. Artificial Intelligence in Healthcare
Artificial intelligence (AI), which refers to the intelligence demonstrated by machines, has become a focus of attention in recent years and has a substantial impact on many aspects of life. Healthcare is no exception and is an important field of interest in AI research and clinical applications. In the past few years, AI-based medical device technologies using machinery learning and deep learning have been rapidly developed, and some medical devices, including those in radiology, have already been approved by the Food and Drug Administration [3]. Furthermore, the development of this technology is expected to accelerate, given the recent rapid advancement of AI technology. In basic medical research, AI is anticipated to make great contributions, such as extracting useful medical information from mass medical data. In addition, AI is expected to play important roles in realizing the current global trends in precision medicine from the perspective of development of drugs and medical devices.
AI can be broadly classified into three categories based on the learning methodology: supervised learning, unsupervised learning, and reinforcement learning approaches. Supervised learning methodology is the most popular approach in the field of medicine. In supervised learning, both inputs and outputs are provided along with associated training data to build a predictive model that computes the output predictions for the inputs. The predictive model is built upon the investigation of correlations between input and output data. For example, in imaging and clinical data analysis, important patient state secrets can be extracted based on the correlation of image data and clinical symptoms. In contrast to supervised learning, unsupervised learning does not take the output data into consideration and applies advanced statistics to purely analyze the structure of input data. Cluster analysis that categorizes various input data into small groups based on the closeness of input patterns is a classic example of unsupervised learning. Reinforcement learning is different from the abovementioned approaches by using reward signals. It basically aims to find inner best strategies to optimize an explicit target task. AI technologies have the potential to greatly alter the landscape of the healthcare industry by expanding the capacity of the workforce and facilitating the delivery of standard services at a reduced cost, while also allowing the redistribution of resources to other important activities [4].
2.1. Overview of AI Technologies
Artificial intelligence (AI) is an umbrella term covering a wide range of algorithms that exhibit human-like intelligence. AI technologies have steadily transformed healthcare delivery, focusing on increasing efficiency, improving patient outcomes, and minimizing costs [1]. Nonetheless, poor implementation of AI applications has impeded its successful penetration in the healthcare setting. Therefore, understanding the stakeholders’ perceptions and varying roles becomes imperative to developing practical solutions and improving implementation strategies fully.
Machine learning algorithms are AI by exploring and processing vast amounts of complex data for patterns and relationships. Such exponential growth in medical data creates hurdles for medical experts who need assistance in analyzing big data. This results in greater interest, focus, and exploration of ML algorithms to advance healthcare delivery. Advances in high-range and low-cost computing paired with thin-client designs have augmented machine efficiency. Thus, ML algorithms are now able to predict the progression of diseases in the early stage of large cohorts of patients employing clinical readouts. Since biomedical data contain a high volume of complex high-dimensional data, it needs to be correctly classified and indexed [5]. An intelligent ML model with supervised, unsupervised, and semi-supervised classifications learns from the given data to conduct learning and categorization of unobserved data. The widely adopted classifiers include linear and non-linear classifiers such as linear regression, support vector machines, naïve Bayes, k-nearest neighbors, decision trees, random forests, neural networks, Gaussian processes, and deep learning algorithms. Multi-layer and convolutional neural networks with deep learning have been successfully employed in healthcare scenarios for image-based diagnostics. Since processed biomedical data is complex and high-dimensional, incremental dimensionality reduction is necessary for a smooth class boundary, which has been explored using manifold learning approaches.
Natural language processing is fundamental to most AI applications in healthcare. For instance, NLP-based algorithms quickly extract valuable knowledge from unstructured notes, which account for almost 80% of clinical data. Manually sifting through such notes is arduous and time-consuming. However, NLP algorithms applied to textual clinical notes can extract data readily for developing new insight. Robotics and automation have been deployed in healthcare processes ranging from pill dispensers for the elderly to robotic surgical assistant systems. In certain cases such as brain and heart trauma surgeries, robots easily outperform human dexterity. With the combined application of AI and ML technologies in healthcare areas from robotics to computer vision being exploited to optimize medical procedures, it has a transformative effect.
3. Applications of AI in Personalized Medicine
Applications of artificial intelligence, focusing on drug discovery and development, and the improvement of treatment regimens. AI in drug discovery and development is enabling the rapid discovery of novel therapeutic targets, while AI in precision medicine and personalized medicine is leveraging patient data to customize treatment plans that inform clinicians of the most effective drugs for each patient.
Using AI-enabled decision support tools could reveal real-world advantages in mitigating adverse effects, increase efficacy, and lower treatment costs. The gradual implementation of AI systems in pharmacogenomics, cancer, and rare diseases would reshape the pharmaceutical industry and clinical practice. The prevailing Precision Medicine (PM) approach determines the most suitable treatment plans for individual patients by taking into account a range of criteria [6]. Some of the analyzed factors include tumor characteristics, treatment regimen history, dose, and drug selection. AI’s role in PM is anticipated to expand, covering critical areas such as diagnosis, drug discovery and development, and patient-matching with drugs based on their specific genomic or genetic alterations. The first attempts to implement AI as part of PM have been undertaken in recent years and showed promising results.
3.1. Drug Discovery and Development
Personalized medicine is an emerging medical model that empowers the clinician to tailor the treatment plan in precision with the specific needs of each individual patient. In recent years, drug discovery and development have experienced two paradigm shifts that have shaped the landscape of personalized medicine technologies. On the one hand, advances in high-throughput experimentally inexpensive biotechnologies have led to the generation of enormous amounts of heterogeneous biological data across diverse scales, from sub-cellular to organism levels. On the other hand, ever-increasing computational capabilities and ever-decreasing hardware costs have enabled new ways of analyzing and interpreting these big data by sophisticated data mining techniques, including machine learning, data assimilation, predictive modeling, network analysis, and sparse representation.
How to efficiently integrate these big biological data with bioinformatics techniques to benefit drug discovery and development have become the key issues. Predictive modeling techniques have been widely applied to explore the quantifiable relation between the chemical structure of a drug and its pharmacological effects by addressing the ADMET issues [7]. In particular, machine learning techniques, included but not limited to kernel-based methods, Bayesian approaches, random forests, and neural networks, have gained some success in ADMET-assisted absorption modeling, toxicity prediction, and bioavailability screening. Guest et al. focused on the recent advances in these AI-driven computational methods and their future perspectives in drug discovery and development. It is necessary to develop an effective and efficient way to properly pre-process biological data so that it is applicable for big data-based predictive modeling [8]. Artificial intelligence-driven big data analytics has the advantage of individually analyzing a data type or multiple types simultaneously.
4. Challenges and Ethical Considerations
The impact of artificial intelligence (AI) on personalized medicine is a complex issue that needs to be understood, managed, and treated carefully. With the growing likelihood of AI-driven healthcare use cases, ongoing ethical debates focus on the transparency and robustness of AI decision-making, health data security, and system abuse [9]. Recently, the US FDA published its AI Action Plan outlining how the FDA intends to regulate AI in healthcare devices. In this plan, the FDA expresses its intent to establish “good machine learning practices” for healthcare devices. Furthermore, the FDA intends to oversee how algorithms behave in real-world health IT environments, develop methods for rooting out bias enabling progressive monitoring of commercial AI, and capping AI business practices at discharge.
With the truly global spread of AI and the lack of an existing comprehensive legal framework governing its global use, ethical, governance, and compliance concerns currently fall squarely in the non-healthcare for-profit realm. Moreover, while the core principals of the European GDPR have been adopted as foundational data privacy principles in many jurisdictions including Brazil and California, there is little agreement on precise definitions and rules that govern the ethical use of the data itself [10]. As a result, a landscape animating compliance with ethics should be anticipated consisting of interpretively rich, ambiguous, and fragmented definitions of responsible AI, thereby holding industry participants accountable according to their context-sensitive interpretations.
4.1. Data Privacy and Security
The rapid advancement of artificial intelligence (AI) in healthcare is the introduction of new AI technologies across a range of healthcare applications. Using AI technologies in healthcare will often involve processing vast amounts of personal health information [10]. There are therefore a range of privacy considerations, risks and concerns raised by implementing healthcare AI, arising from both the use of for-profit businesses and the commodified nature of the health information. There are also challenges for protecting this information in a decentralized AI-enabling environment comprised of independent parties (e.g., research institutions, hospitals, healthcare providers, technology companies) using a mix of proprietary technologies, interoperable systems and open-source algorithms. New parties, arrangements, technologies and devices are expected to emerge within the evolving healthcare AI landscape, further complicating and tightening control over health information.
Healthcare AI also raises ethical issues concerning the control, employment and use of health information [11]. Health data can be misused for purposes other than those for which it was intended. Decisions regarding data access, use and policy—concerning a valuable public asset common to all—that shape the commercial future of healthcare systems should not be determined solely in the interests of private enterprise. AI technologies are also capable of making decisions that affect patients’ health outcomes, determining the nature of their treatment, or approving/denying insurance claims. Such AI systems could shape the course of people’s lives in significant ways and, because they can be kept secret, undermine the right to a rationale explaining why a certain decision has been made.
5. Future Directions and Opportunities
The final section examines future directions and opportunities within the field of personalized medicine and artificial intelligence. Despite the recent interest in implementing AI, there is minimal real-world evidence regarding how predictive analytics impacts patient outcomes, subsequently informing areas of improvement and optimization [5]. However, AI has the potential to increase the accessibility of personalized medicine. As better algorithms, mathematics, and computational ability continue to be developed, more commonplace screening could occur at the population level. To augment current models of personalized medicine, AI-driven predictive models should be integrated into the clinician-patient relationship, where collected patient information would flow into AI systems. These intelligent systems would then assess the current disease from a multi-omics perspective, offering alternatives for diagnosis and treatment [1].
Discovering novel biomarkers for diagnosis could include machine vision/image analysis, protein-array analysis, and hybrid approaches combining different sources of data. Finding biomarker signatures could draw on connections with other “omics” disciplines, such as low-resolution structure-based genomics. Such discoveries could occur at many possible scales, from localities relevant to the body framework to large hoods of great interest in particular disease types. Once a signature or pool of identified candidate biomarkers is achieved, AI systems might augment the development of diagnostic tests based on bioassay platforms and detection methods or technologies. For diagnosis, tests could be designed to assess different biological levels (proteomics/genomics), providing unique information and large discrepancies for each context of the body. For treatment, platform designs might include novel treatments with profiling involvement and monitoring of the body-cell signaling network. Given the potential consequences of non-appropriate choices, the thorough complement of carefully selected alternative prevention or treatment measures is needed. Crucially, this needs to build trust in how AI systems reach specific conclusions, insights, and recommendations through intelligent and socially-consistent processes. Consequently, as systems update recommendations and treatment choices, it is crucial to continue feeding back data routinely on whether results from AI systems met expectations.
As AI systems become routine in the personalized medicine endeavor, new opportunities would emerge for collaborative works involving researchers and engineers with backgrounds in mathematics, physics, and computer science. Data from clinical trials would connect clinical and AI experts, while interactions with other disciplines would index imbalances between the maximal potential of precision medicine and its realistic implementation. Models of personalized medicine augmented by AI have the potential to bring great improvements and become a prominent trend in both contemporary and future medicine.
References:
[1] Z. Ahmed, K. Mohamed, S. Zeeshan, and X. Q. Dong, “Artificial intelligence with multi-functional machine learning platform development for better healthcare and precision medicine,” 2020. ncbi.nlm.nih.gov
[2] W. Evans, E. M. Meslin, J. Kai, and N. Qureshi, “Precision Medicine—Are We There Yet? A Narrative Review of Precision Medicine’s Applicability in Primary Care,” 2024. ncbi.nlm.nih.gov
[3] R. Hamamoto, K. Suvarna, M. Yamada, K. Kobayashi et al., “Application of Artificial Intelligence Technology in Oncology: Towards the Establishment of Precision Medicine,” 2020. ncbi.nlm.nih.gov
[4] S. Wójcik, A. Rulkiewicz, P. Pruszczyk, W. Lisik et al., “Beyond ChatGPT: What does GPT-4 add to healthcare? The dawn of a new era,” 2023. ncbi.nlm.nih.gov
[5] D. Dixon, H. Sattar, N. Moros, S. Reddy Kesireddy et al., “Unveiling the Influence of AI Predictive Analytics on Patient Outcomes: A Comprehensive Narrative Review,” 2024. ncbi.nlm.nih.gov
[6] C. Carini and A. A. Seyhan, “Tribulations and future opportunities for artificial intelligence in precision medicine,” 2024. ncbi.nlm.nih.gov
[7] W. Chen, X. Liu, S. Zhang, and S. Chen, “Artificial intelligence for drug discovery: Resources, methods, and applications,” 2023. ncbi.nlm.nih.gov
[8] A. Ioana Visan and I. Negut, “Integrating Artificial Intelligence for Drug Discovery in the Context of Revolutionizing Drug Delivery,” 2024. ncbi.nlm.nih.gov
[9] S. Pasricha, “AI Ethics in Smart Healthcare,” 2022. [PDF]
[10] B. Murdoch, “Privacy and artificial intelligence: challenges for protecting health information in a new era,” 2021. ncbi.nlm.nih.gov
[11] A. Oliva, S. Grassi, G. Vetrugno, R. Rossi et al., “Management of Medico-Legal Risks in Digital Health Era: A Scoping Review,” 2022. ncbi.nlm.nih.gov