Hearing the Full Orchestra: Multiomics, AI, and the Future of Therapeutic Discovery

A genome can reveal the biological instructions a person inherited. It cannot, by itself, show which instructions are active, how much protein a cell is producing, or how disease, treatment, environment, and time are altering those processes.

That limitation helps explain the growing interest in multiomics, an approach that integrates multiple layers of biological information to develop a more complete picture of health and disease. With the aid of machine learning and other analytical tools, multiomics may help researchers identify disease subtypes, discover biomarkers, prioritize drug targets, predict treatment response, and find new uses for existing therapies. The challenge is turning increasingly complex data into conclusions that are reproducible, clinically meaningful, and sufficiently reliable to guide decisions.

From One Instrument to the Full Orchestra

The term “omics” refers to the large-scale study of a particular category of biological information. Common examples include:

• Genomics, the study of DNA and genetic variation;
• Epigenomics, the study of chemical and structural changes that influence gene activity without changing the underlying DNA sequence;
• Transcriptomics, the study of RNA molecules produced when genes are expressed;
• Proteomics, the study of proteins produced by cells and tissues; and
• Metabolomics, the study of small molecules generated through cellular metabolism.

Each layer provides a different type of information. Genomics may identify a variant associated with disease risk, while transcriptomics can show whether a related gene is active in a particular tissue. Proteomics can indicate whether the corresponding protein is present, and metabolomics can reveal downstream changes in cellular activity. 

Integrating these layers may expose relationships that are difficult to see when each dataset is analyzed independently. The National Institutes of Health describes multiomics as a way to develop a more holistic view of the molecular factors and cellular processes involved in health and disease. NIH identifies disease classification, biomarker discovery, and drug-target identification among the approach’s potential applications.

Studying one omics layer is like listening closely to one section of an orchestra. The information may be valuable, but it does not necessarily reveal how the full composition works. Multiomics attempts to understand how different biological layers interact and which signals may be driving disease or treatment response.

Genomics Was a Beginning, Not an Endpoint

The international Human Genome Project, launched in 1990 and completed in April 2003, produced the first reference sequence of the human genome and accelerated research into human biology and disease. It was a landmark achievement, but it did not completely explain how genes produce health or disease.

Genome-wide studies have since identified many associations between genetic variants and diseases or other traits. An association, however, does not necessarily establish causation, identify the cell type through which a variant acts, or show that changing the associated biological target will improve a clinical outcome.

Multiomics can help investigate these questions by connecting genetic evidence with gene expression, proteins, metabolites, cell types, clinical characteristics, and environmental exposures. The approach does not eliminate uncertainty, but it provides additional evidence with which to evaluate it.

Applications in Therapeutic Discovery

Multiomics may contribute at several stages of drug development.

- Identifying Drug Targets and Patient Populations

Integrating different omics layers can help researchers identify consistently disrupted biological pathways. A target supported by genetic, transcriptional, protein, and metabolic findings may warrant greater attention than one supported by a single association.

Multiomics can also identify the cell types or disease states in which a target appears relevant. That context matters because a protein associated with disease may perform beneficial functions elsewhere in the body, creating safety concerns if it is inhibited too broadly.

Similarly, one diagnosis may encompass several molecularly distinct forms of disease. Multiomics may help classify patients by biological characteristics rather than symptoms alone, potentially improving clinical-trial design and identifying the people most likely to benefit from a therapy.

Reviews of multiomics in therapeutic research describe potential applications in target identification, disease classification, biomarker discovery, and precision treatment. Computational findings still require experimental and clinical validation.

The NIH Multi-Omics for Health and Disease Consortium is examining integrated omics, phenotypic, environmental, and social determinants of health data in ancestrally diverse populations. The consortium is also developing approaches to data harmonization, integration, standards, and controlled access.

- Developing Biomarkers and Repurposing Drugs

Multiomics may support biomarkers used to detect disease, estimate prognosis, select patients for treatment, or measure biological response. But finding a molecular signature in a dataset is not the same as establishing a validated biomarker. The analytical method must be reliable, and the biomarker must be evaluated for its intended context of use.

In April 2026, FDA issued final guidance addressing the validation of bioanalytical methods used to measure biomarker concentrations in biological matrices such as blood or urine. The guidance is nonbinding, but it illustrates the importance of reliable analytical methods when biomarker concentration data are used in regulated drug-development programs.

Researchers are also using multiomics to explore new indications for existing drugs. Because information may already exist about a drug’s safety, manufacturing, and pharmacology, repurposing can sometimes reduce portions of the development process. One peer-reviewed computational study used multiomics data and deep learning to predict cancer drug response and identify potential repurposing opportunities. Such predictions remain hypotheses rather than proof that a drug will be effective for a new indication.

The Role of Artificial Intelligence

Multiomics datasets can contain millions of variables across relatively small participant populations. Machine learning and other forms of artificial intelligence can assist with pattern recognition, data integration, patient stratification, and prediction. Researchers may also combine omics information with medical imaging, electronic health records, and digital-health or real-world data.

But more data do not automatically produce better evidence. AI models can amplify weaknesses such as missing values, inconsistent laboratory methods, population imbalance, batch effects, and mislabeled outcomes. A model may perform well during development but fail when applied to another institution or patient population. Its performance may also deteriorate as data and clinical practices change over time.

In its January 2025 draft guidance, FDA proposed a risk-based credibility framework for AI models used to generate information supporting regulatory decisions about drug safety, effectiveness, or quality. The draft remains nonbinding and not final, but it emphasizes that model credibility depends on its particular context of use and the consequences of an incorrect result.

AI-generated multiomics findings should therefore remain subject to scientific review, independent validation, documented data lineage, version control, and appropriate human oversight.

Data, Privacy, and Operational Risk

Multiomics may require companies to collect, combine, store, and share sensitive biological and health information. Genomic data create particular privacy concerns because DNA is inherently connected to an individual and may reveal information relevant to biological relatives. The National Human Genome Research Institute notes that genomic data can retain reidentification risk even when conventional identifiers have been removed.

Applicable obligations depend on the organization’s role, research activities, data sources, contractual commitments, and jurisdictions. Relevant considerations may include:

• Participant consent and permitted secondary uses; 
• Institutional-review and protocol requirements; 
• The Common Rule, where applicable; 
• HIPAA, where protected health information is used or disclosed by a covered entity or business associate; 
• State genetic- and health-data privacy laws; 
• NIH data-sharing requirements for funded research; 
• International data-transfer restrictions; 
• Cybersecurity and access controls; 
• Reidentification and downstream-sharing restrictions; and 
• Data quality, bias, representation, and sample-processing differences.

For NIH-supported projects generating covered genomic data, the NIH Genomic Data Sharing Policy establishes expectations concerning data-management plans, institutional certifications, timely repository submission, and responsible use of controlled-access data.

The Global Alliance for Genomics and Health has also developed frameworks and policy resources for responsible genomic and health-data sharing. These resources are not substitutes for applicable law, but they can help inform data-governance programs and contractual expectations.

Hearing the Full Performance

Multiomics expands the questions researchers can ask. Instead of examining DNA, RNA, proteins, or metabolites in isolation, they can study how those layers interact across cell types, patient populations, environments, and time.

The approach may improve disease classification, target selection, biomarker development, clinical-trial design, and treatment prediction. Its value, however, will depend less on the volume of data collected than on the quality, representativeness, governance, and validation of that data.

Listening to more of the biological orchestra can reveal patterns that one instrument cannot. The next challenge is determining which patterns are reproducible, which are clinically meaningful, and which can be translated responsibly into safer and more effective therapies.

Authored by Katie Vaitkus, AVP, Territory Manager, Berkley Life Sciences