Alone, even the most comprehensive omics dataset offers an incomplete picture of human biology. Multiomics, which combines multiple levels of biology, including genomics, transcriptomics, proteomics and metabolomics, provides researchers with a more holistic view of a biological system. This system-level view is essential for understanding disease mechanisms and supporting more rational drug design.
However, despite its promise, the clinical impact of multiomics has yet to be fully realized. “In many cases, multiomics data includes data pieced together from various places,” Prof. Namshik Han, professor in the Department of Quantum Information at Yonsei University, head of AI research at the Milner Therapeutics Institute and chief technology officer and co-founder of CardiaTec Biosciences, told Technology Networks at the ELRIG Drug Discovery 2025 event. “For example, you might have gene expression data from one cohort in the US, DNA sequencing from a second cohort in Africa and epigenetic data from a third cohort in Japan.” This fragmented approach makes data integration complex and unreliable, which poses a significant barrier to its application in drug discovery.
To overcome this, Han and his colleagues at CardiaTec are generating high-quality, unbiased multiomic data to build the largest multiomics database of human cardiac tissue. “What we have set out to do is generate multiomics data from the same tissue in a single patient,” said Han.
AI and multiomics: A match made in heaven?
AI has transformed the drug discovery and development process, allowing developers to model drug-disease interactions, predict efficacy and toxicity and optimize compounds in silico before a molecule ever reaches the lab.
When paired with multiomics, AI can transform biological complexity into structured, predictive frameworks. Not only can AI work with the high-dimensional data typical of omics studies, but AI models also stand to benefit from being trained on the complementary information contained in multiomics modalities.
The application of AI in multiomics also holds promise in the development of personalized medicine, where deep learning could be used in combination with multiomics data to analyze and predict personalized drug responses. However, before this can be achieved, AI models need to be trained on population-based data. Han explained, “To provide context for personalized medicine, we need to be able to compare results with a larger population, which may take time, as we need a huge amount of data.”
A key factor in ensuring the future success of AI in multiomics data interpretation is the training of the models themselves. “AI models require a large amount of data for training. Even more importantly, it requires high-quality data without which it will produce very poor predictive results,” said Han.
Fragmented or inconsistent data are often associated with multiomics, which even the most advanced algorithms will struggle to use to learn valuable patterns. It is for this reason that many early AI applications have failed to deliver hypothesis-generating insights.
Solving the data fragmentation challenge requires the development of next-generation multiomics platforms that can capture multiple forms of data from a single sample in a single experimental run alongside computational pipelines that can integrate multi-modal data.
Integrated multiomics is already providing insights into diseases that would be nearly impossible to obtain through a single-modality analysis alone. For example, researchers integrated genomic, transcriptomic and proteomic characterization of endometrial carcinomas, revealing the existence of distinct molecular subtypes of pancreatic cancer.
When combined with the power of AI, integrated multiomics can be used to move beyond identifying patterns, to revealing unexpected correlations and generating new hypotheses that can be tested in vitro.
Education is key to unlocking AI’s potential in multiomics
Fully realizing the potential of AI in multiomics is not just a technological endeavor, but one that requires human expertise. “The biggest hurdle is not developing an AI algorithm, but having a biological understanding of the multiomics data, because without that, we cannot create a functional algorithm,” explained Han. “In many ways, we need to work together across multiple disciplines. Ideally, we want researchers who have both biological and computer science knowledge, so that they can understand the data first and develop or refine an algorithm using that expertise.”
This knowledge base is not only vital when creating AI models to analyze the data, but also plays a role in being able to interpret and explain what a computational prediction means biologically. “Without that, a prediction is just a prediction,” stated Han.
A barrier to fully realizing the potential of AI in drug discovery may come from a lack of AI expertise in the healthcare domain, which Han explains has arisen due to the competitiveness of the tech industry. “Computer science graduates can go into many other sectors, and maybe some of them are more attractive because they offer a higher salary. On the promising side, many students can see the value of utilizing AI in drug discovery to ultimately save lives and so are choosing this path.”
“What we can do is educate young students to become the next generation of AI healthcare researchers. Additionally, we need to upskill current researchers in industry or academia to allow them to understand and utilize AI to its fullest,” Han concluded.
By pairing multiomics with AI, it is possible to uncover new patterns in high-dimensional, multiomics datasets that are beyond human capability to discern. The synergy of AI and multiomics represents a shift from static snapshots of biology to dynamic models of disease that can accelerate the discovery of transformative therapies. Consistent, scalable data and human expertise will play vital roles in this next chapter in drug discovery.
