A cornerstone of precision medicine is the technological platforms enabling unprecedented volume and insights in combined data. This section will highlight the existing and upcoming platforms with recent notable examples of clinical application, while a comprehensive overview has been described elsewhere.¹⁶ Next-generation sequencing (NGS) allows massive parallel sequencing of short DNA sequences in a ‘high-throughput’ format and then assembling information.¹⁷ Over the past decade, NGS has accelerated academic discoveries and is now translating to the clinic as an actionable tool for patient care. WGS has aided the discovery of disease-causative or associated gene variants in rare diseases, somatic variant calling in cancer, and predicting polygenic risk scores for genome-wide association studies (GWAS) that have spurred direct-to-customer genetic testing platforms.¹⁸ A comprehensively equivalent data-enriched platform and economic alternative is whole-exome sequencing (WES), which involves targeted sequencing of only the coding regions of the genome. This has led to the FDA approval of two test panels currently available for diagnosing and supporting treatment regimens by matching cancer patients (solid tumors) to clinical trials.^19,20 To overcome limitations of WGS/WES that do not address the impact of aberrant gene expression and alternative gene splicing and to complement/improve the accuracy of diagnosis through genomic approaches, RNA (coding, non-coding, regulatory) sequencing or transcriptomics was developed.²¹ Clinical use of RNA sequencing has improved the diagnosis of Mendelian diseases²² and hematological malignancies.²³ NGS has supported treatment/care management decisions or predicting cardiovascular, neurogenetic, and neurological disease susceptibility.^24–26 Infectious disease surveillance and management has benefited from NGS and RNA-Seq platforms,^27,28 which use the long read or deep-sequencing approach to facilitate host–pathogen genetic distinction.²⁹

NGS provides a wealth of data, albeit ‘bulky,’ which underappreciates the complexity of disease states and their heterogeneity, something paradoxical to precision medicine. Analytical platforms at the single-cell and single-nucleus levels have been developed over the past decade to bridge this knowledge gap. These interrogate the genome (single-cell genome-sequencing),³⁰ the transcriptome (single-cell RNA sequencing)^31–33, and the proteome (single-cell proteome sequencing),^34,35 and in a truly multi-modal methodology is integrating spatial information in tissues to understand the complexity of cellular organization that can provide insights into disease states.^36,37 The miniaturization and high throughput of single-cell technologies were enabled by microfluidics, a technique using parallel microchambers on specially fabricated material also known as ‘lab-on-a-chip.’³⁸ Whole-genome application using droplet-based microfluidics platforms allows the identification of single nucleotide variants at the single-cell level. Similarly, droplet-based massively parallel RNA sequencing methodologies have resulted in cost-effectiveness and high throughput using the 10X Genomics Chromium and Drop-Seq platforms.^39–43 These are integrated to understand the epigenome (e.g., methylation) using single-cell bisulfite sequencing, chromatin assembly (transposase-accessible chromatic sequencing, ATAC-seq), and single-cell protein repertoire investigations using mass spectrometry methods such as matrix-assisted laser desorption/ionization to provide a holistic picture of the individual cellular organization.⁴⁴

Although many of these have yet to be translated into the clinical setting, notable use cases are in the field of oncology due to the inherent complexity of cancer. Liquid biopsy (using blood samples) has gained traction for the detection of circulating tumor cells, cell-free RNA, circulating tumor DNA, and protein biomarkers for detecting minimal residual disease using WGS/WES in cancer patients.^45,46 Single-cell RNA sequencing is now used for patient stratification for drug response and biomarker discovery and aiding in drug discovery in oncology.⁴⁷ Precision medicine has enabled the approval of 11 gene therapies between the US FDA and the European Medical Association.⁴⁸ Gene editing tools such as CRISPR/Cas9 (clustered regularly interspaced palindromic repeats) have contributed to precision medicine in the clinic with the most recent FDA approval of a therapy for sickle cell disease.⁴⁹ Viral or non-viral gene delivery methods enabling the approval of chimeric antigen receptor T cell therapies, Tisagenlecleucel and axicabtagene ciloleucel,^50,51 and the next generation of these platforms are progressively moving to the clinic.⁵²

Genomic medicine has been used to limit the ‘one-size-fits-all approach’ by considering that knowledge of gene–drug interactions can better tailor and predict a patient’s chances of managing their treatment regimen and curtail any potential adverse events resulting from combination therapeutic regimens.^53,54 For instance, advances in pharmacogenomics have enabled the understanding of the COVID-19 disease spectrum for susceptibility and response to existing drugs.⁵⁵ Another key area where precision medicine is impacting patient care is the field of ‘radiomics.’ Traditionally, ‘radiomics’ is defined as “the high-throughput mining of quantitative image features from standard-of-care medical imaging that enables data to be extracted and applied within clinical decision support systems to improve diagnostic, prognostic, and predictive accuracy”.⁵⁶ Using high-quality imaging techniques and integrating patient disease information to develop statistical modeling, ‘radiomics’ aims to develop a biomarker-guided approach for precision disease prognosis/diagnosis.⁵⁷

Particularly, in oncology, supplementing genomic and molecular information from tumor heterogeneity with imaging phenotypes has enabled patient stratification for better disease management. Finally, clinical studies are being redesigned with an underpinning of using disease biology and potential biomarkers across therapeutic indications to improve real-time clinical decisions to maximize benefit for the patients.⁵⁸

Precision medicine implementation presents several problems that call for an all-encompassing strategy that includes regulatory changes, technological developments, collaboration, and educational programs. Integrating various data sources used in precision medicine requires standardizing data formats and enhancing interoperability across various healthcare systems. The absence of established data formats and interoperable systems makes it difficult to integrate different types of data (such as genetic, clinical, and environmental). Data fragmentation can be avoided by taking steps like creating common data models and following global standards like HL7 FHIR (Fast Healthcare Interoperability Resources).⁵⁹

Many strategies, including differential privacy, blockchain for safe data sharing, and sophisticated encryption mechanisms, have been developed to protect genomic data. While allowing its use in research and individualized care, these strategies seek to safeguard sensitive data.⁶⁰ Significant financial obstacles are created by the high expenses of genome sequencing, data processing, and customized treatments, especially in environments with limited resources. Reducing costs can be achieved by implementing sequencing technologies that are cost-effective, encouraging value-based care models, and cultivating public–private collaborations. Enhancing access can also be achieved by increasing insurance coverage for customized treatments and genomic testing.⁶¹

The regulatory environment around precision medicine is complicated, with different requirements and policies governing the acceptance of customized treatments and diagnostics. It is essential to create precise regulatory frameworks and rules that handle the difficulties presented by precision medicine. To develop standards that are both rigorous and adaptable, regulatory authorities must collaborate extensively with researchers and industry stakeholders. Healthcare professionals, many of whom lack training in genomics or tailored care, must undergo substantial education in addition to considerable modifications in clinical workflows to successfully incorporate precision medicine into clinical practice. It is crucial to provide focused educational initiatives and ongoing professional development programs for healthcare providers.⁶²

Several ethical and social concerns are brought up by precision medicine, such as those related to equity, access to care, genetic prejudice, and the possibility of making health inequities worse. It is crucial to create moral guidelines and regulations that handle these issues. Initiatives for public participation and education can contribute to the development of trust and guarantee that the advantages of precision medicine are shared fairly.

To involve patients in precision medicine, it is necessary for them to comprehend intricate genetic data and the consequences of customized treatment alternatives. Insufficient health literacy may impede patients’ involvement. Patient participation can be improved by streamlining communication, educating patients with digital technologies, and developing materials that clearly and concisely outline the advantages and disadvantages of precision medicine.

Precision medicine generates enormous amounts of data, including multi-omic and genomic data, which are difficult to interpret without specialized knowledge and advanced tools. Clinicians can better understand data and make well-informed treatment decisions by utilizing AI and machine learning (ML) in sophisticated clinical decision support systems.

Significant obstacles are presented by worries about the security and privacy of genomic data, including the possibility of re-identification and unlawful access. Particularly in low-resource areas, access may be restricted by the high costs of genome sequencing, data analysis, and individualized medicines. The approval of tailored medicines and diagnostics is subject to varied requirements within the complicated regulatory environment of precision medicine. Changes in clinical processes and healthcare personnel’s education are necessary to incorporate precision medicine into clinical practice, as many of them lack genomics training. Artificial intelligence (AI) can greatly aid in overcoming the obstacles that precision medicine now faces in all of this.

Neural networks having an input layer, one or more ‘hidden’ layers, and an output layer are used in traditional ML techniques. These algorithms are typically limited to supervised learning, which means that for the algorithm to find characteristics in the data, human specialists must arrange or annotate the data. Three steps are involved in the ML process: decision--making, error evaluation, and model optimization. Based on the input data, the algorithm generates predictions, assesses its accuracy, and iteratively modifies weights to increase accuracy. ML models include the following:

Multiple layers of nodes (neurons) are used in deep learning to bridge the network’s inputs and outputs. From the unprocessed input, these layers can gradually extract higher-level information. In image processing, for example, lower layers might detect lines, while higher layers might detect more complicated ideas like faces, letters, or numbers.^67–69 Multiple layers of interconnected nodes make up deep neural networks, and each layer builds on the one before it to improve and optimize the classifications or predictions. We refer to this series of calculations throughout the network as forward propagation.

In a different procedure known as backpropagation, prediction errors are computed using techniques such as gradient descent. The function’s weights and biases are then modified by going back through the layers to train the model. A neural network can make predictions and fix mistakes when it uses both forward propagation and backpropagation together. A large amount of processing power is needed for deep learning. The best graphics processing units are those with high performance, as they have many cores and enough memory to execute heavy computations. Dispersed cloud computing could be beneficial as well. Deep learning requires this kind of processing power in order to train deep algorithms.

The algorithm gets increasingly more precise over time. Convolutional neural networks (CNNs) and recurrent relational networks (RNNs) are two examples of deep learning models that are now applicable in the medical industry. Applications related to computer vision and image categorization are the primary uses for CNNs. They can carry out tasks like object identification, image recognition, pattern recognition, and face recognition because they can identify characteristics and patterns in pictures and videos. Conversely, recurrent neural networks, or RNNs, are typically employed in speech and natural language recognition applications. Their feedback loops set them apart and are intended for use with sequential or time-series data. When working with time-series data to anticipate future outcomes, these learning techniques are frequently used.

Combining rule-based modeling of human language with statistical modeling, ML, and deep learning, natural language processing (NLP) allows computers and digital devices to detect, comprehend, and produce text and speech. NLP is a combination of deep learning, ML, and computational linguistics. A subfield of linguistics called computational linguistics analyzes speech and language using data science. Semantic analysis and syntactic analysis are the two primary forms of analysis involved. Syntactic analysis uses preprogrammed grammar rules to parse word syntax and determine meaning for words, phrases, and sentences. Semantic analysis interprets the meaning of the words inside the sentence structure by using the syntactic analysis output to extract meaning from the words.

Given that the possibilities for enhancing medicine with AI in its earliest version have increased exponentially over the past decade, several notable use cases exist.⁷⁰

High-quality, varied, and well-integrated datasets are necessary for AI models; however, inconsistent, fragmented, and incomplete medical data are frequently encountered. Strong data cleaning and preprocessing approaches are necessary to ensure that the data supplied into AI models is reliable and correct. Federated learning techniques preserve privacy and increase data diversity by enabling AI models to learn from distributed data that is dispersed across multiple sites without the need for data centralization. Standardized EHRs and data sharing initiatives like the NIH’s All of Us Research Program may enhance data integration and quality.

Due to AI models’ propensity to reinforce biases included in training data, different population groups may be treated differently. A key component of bias mitigation is making sure that training datasets are representative of different demographic groups and are varied. Clinicians’ acceptance and trust of AI models can be increased by the development of explainable AI techniques that provide insights into the models’ decision-making process.⁸⁵

The creation of ethics committees to examine AI applications in the healthcare industry can be a useful way to resolve ethical challenges related to equity, consent, and data ownership. The General Data Protection Regulation of the European Union and the AI Act offer frameworks that may direct the moral and legal application of AI in precision medicine.

Collaborative efforts between public health agencies and private enterprises hold the potential to alleviate the financial burden on healthcare providers. Initiatives like the Global Health Initiative are dedicated to rendering AI-driven precision medicine accessible to under-resourced regions.⁸⁶ AI has promising potential for improving the outreach and impact of precision medicine in the future (Figure 1).