June 18, 2025

Combination therapy using multiple drugs is currently being used in cancer treatment to reduce the risks of side effects and increase the therapeutic effects. In addition, "drug repositioning," which applies existing drugs to other diseases, is attracting attention in the pharmaceutical industry from the viewpoint of cost optimization and recovery in new drug development.

Under these circumstances, NEC announced a demonstration experiment of its "Graph-based AI Technology for Predicting Drug Combinations for the Enhanced Efficacy of Cancer Treatment" to suggest new drug combinations that can be used in combination therapy for cancer. We conversed with several researchers to learn more about this technology which suggests new possibilities for cancer treatment while simultaneously enabling the effective use of existing drugs.

Suggesting unknown drug combinations that are effective at treating cancer together with convincing evidence

―Please tell us more about this recently announced "Graph-based AI Technology for Predicting Drug Combinations for the Enhanced Efficacy of Cancer Treatment."

Ohnishi: It is an AI base technology to suggest candidates for new drug combinations in cancer combination therapy. However, it not only suggests candidates but also presents convincing evidence. In cancer treatment, multiple drugs are often used in combination to reduce the impact of side effects and increase the drug efficacy. However, many patients are unable to achieve satisfactory results with this approach. The purpose of this technology is to provide such patients with effective treatments at an early stage by helping researchers explore new drug combinations.


Kubo: We can expect that drugs that have not been used in conventional cancer treatments will be suggested as combination candidates. For example, there are cases where the indications of anticancer drugs approved for use in breast cancer and stomach cancer have been expanded to treat salivary gland cancer, so a variety of approaches can be envisioned such as combinations that expand the drug indications to different types of cancer.

In recent years, the drug repositioning movement has been gaining momentum in the pharmaceutical industry. This movement tries to discover existing therapeutic agents that may be effective at treating other diseases. Easy-to-understand examples of this trend include aspirin, which was developed as an antipyretic analgesic and later used as a thrombolytic agent, or minoxidil, which was developed to treat high blood pressure and became used to promote hair growth.

Creating a new drug typically requires ten or more years of research and development plus five to seven years of clinical trials. Nevertheless, even if such costs are incurred, less than 5% of new drugs make it to the approval stage. Furthermore, with drugs having been almost thoroughly developed for treating diseases that affect many people, pharmaceutical companies must raise their research and development efficiency to even higher levels than ever before.

In response, drug repositioning, which expands the indications of previously approved drugs to other diseases, can significantly shorten not only the research and development period but also clinical trials and other approval processes. It is an important measure for more effectively applying approved drugs that are like precious jewels created at great cost. This technology is also compatible with that trend and is very useful in discovering the potential to expand existing drugs to other indications.


Yano: We developed this technology through collaborative research with Chugai Pharmaceutical. Chugai Pharmaceutical is already using the demo version and has provided positive feedback. In particular, I think that we expect AI to suggest novel ideas that humans would never dream of, and the manager in charge actually praised the technology for producing several new drug candidates. Furthermore, the technology also outputs easy-to-understand evidence to explain its suggestions, which reportedly made the results extremely convincing.

AI utilizes graph data to output unknown combinations with high accuracy

―How did you realize this technology?

Yano: We utilized graph-based AI technology developed by NEC. This AI technology is equipped with a link prediction function that learns from the graph data and can predict new connections between pieces of data. Graph data is a data format that expresses the connections between pieces of data and is composed of dots (nodes) and lines (links) connecting them like a network diagram. It is well-suited to the biological data used when developing drugs. For example, it can express the connections between a particular protein and an effective drug. We used this feature to build a complex graph and constructed a system that suggests new combination drug candidates.

While there are other types of AI that use graph data, the most prominent feature of our technology is that it can present the basis for its predictions. As Kubo mentioned a moment ago, drug development is an enormously expensive process. For that reason, no matter how many new possibilities an AI might suggest, it is impossible to embark on clinical trials without a compelling reason. In fact, Chugai Pharmaceutical also stressed the explainability of AI during the development.

Ohnishi: Another advantage of our technology is the high accuracy of its output. In fact, Chugai Pharmaceutical has been also attempting to utilize their graph data from before starting the full-scale collaborative research, so we established the target accuracy based on their experience. Our technology was able to exceed this target accuracy, allowing the collaborative research moved forward positively.


Yano: The reason for the high accuracy lies in the fact that the NEC AI which handles the graph data combines multiple AI engines. This method was developed by NEC Laboratories Europe and is based on combining AI algorithms cultivated through many years of graph data research and our unique know-how in generating explanations for prediction results.

In this research, we improved the prediction accuracy by combining an engine that predicts new connections to ensure accuracy with an engine that learns the connection rules to create the explainability such as if A, then B and if B, then C.

Ohnishi: In addition, it was also extremely important to consider how to build the graph data in achieving the accuracy. Since expert knowledge of medicine is also required, we built the graph through repeated discussions with Chugai Pharmaceutical and trained the AI on it.

Kubo: Our team also includes researchers with expertise in the areas of medicine and biology who have experience with drug discovery and who have medical degrees, and we have a long history of experience with data science in the medical field. Such expertise may have helped in the research and development.


Yano: Furthermore, the research paper regarding this technology has been accepted and highly evaluated by several top, international conferences including The Conference on Uncertainty in Artificial Intelligence (2018) (Note 1) and the AAAI Conference on Artificial Intelligence (2019) (Note 2).

Aiming to build a massive graph data set utilizing NEC's business foundation and starting new businesses

―Tell us about the future prospects of this technology.

Yano: As a researcher, I would like to explore the possibilities of graph data. As mentioned earlier, graph data is extremely compatible with the biological area. There is definitely room for further application in areas such as developing new drugs and predicting the proteins involved in specific diseases, etc. I think that the predictions of drug combinations at this time are only a subset of such possibilities, so I would like to work on expanding the scope of application of this technology.


Kubo: That's a good question. In that sense, if we can build graph data which is even larger than the one that we have currently built, then we should be able to make more accurate predictions and apply the technology to a broader range of tasks. For example, NEC has the number two share of the electronic medical record business for large hospitals in Japan, and we might be able to link the graph data with such medical data. In addition, the NEC Group has begun working on omics analysis (Note 3) services such as "Fones Visuas" and "BostonGene," and combining these technologies will open up further possibilities. In this way, NEC is a globally rare company with a business foundation that is highly compatible with graph data analysis. We believe that building a massive graph from such a foundation can create highly accurate predictions and task processing that will create new forms of value.


Ohnishi: When it comes to AI, people generally think of image processing and natural language processing that handle image and text data, etc. However, AI that handles graph data which can show the similarities and relationships between data still has significant potential. First, we would like to increase the recognition of AI technologies which handle that graph data. For NEC, we would like to establish an AI team that is dedicated to graph data going forward and increase our presence in order to be able to play an active role in the pharmaceutical and medical areas not only in Japan but globally.

Go back to Featured Technologies