Why Rare Cancers Are Often Left Behind In Drug Discovery

By Paul Romness, chair, chief executive officer, president, OS Therapies

Artificial intelligence is playing an increasingly visible role in drug discovery. Across the pharmaceutical and biotechnology industries, AI tools are being used to analyze biological data sets, identify potential drug targets, and help speed up early stages of development. Many companies are exploring how these technologies can shorten timelines and potentially lower the cost of introducing new therapies to market.

Oncology is seeing a surge in AI adoption. Researchers are now using these tools to unlock insights from genetic and clinical data that were previously too dense or fragmented to analyze effectively. As a result, there are new opportunities for understanding cancer biology and identifying potential treatment strategies. However, while AI-driven discovery is starting to influence some areas of oncology, it also highlights important limitations. Most notably, rare cancers are often missing from the data sets these systems rely on.

As a result, researchers and companies focused on rare cancers continue to see a gap between technological promise and real-world impact. This gap is becoming more noticeable as AI becomes more widely used across oncology research.

Where AI Is Already Changing Drug Discovery

The interest in AI across the pharmaceutical industry is easy to understand. Drug discovery generates large amounts of data, from genomic sequencing to laboratory screening and clinical trial results. AI systems are particularly useful for identifying patterns within these data sets.

Machine learning models are now being used in several parts of the discovery process. In early research, AI can help identify potential drug targets by analyzing genetic changes associated with the disease. These tools also can help predict how certain drug candidates might behave, which allows researchers to focus on the most promising compounds before moving into laboratory testing.

In some cases, AI tools are also being applied earlier in the development process to help evaluate potential drug candidates. By predicting how certain compounds might behave, these tools may help researchers focus their efforts on the most promising options. This could reduce the time and resources required during early development.

These models depend heavily on access to large data sets, and the quality and amount of training data often determine how well an AI system works. This creates challenges for diseases that do not generate large amounts of data.

The Unique Challenges Of Rare Cancers

Rare cancers represent a significant portion of oncology diagnoses, yet each disease affects relatively small patient populations. In many cases, researchers might only see a few hundred or a few thousand cases per year in a given region.

Osteosarcoma, for example, is the most common primary bone cancer but remains extremely rare overall. Approximately 1,000 cases are diagnosed annually in the United States, most often in adolescents and young adults. Despite decades of advances in oncology, treatment approaches for osteosarcoma have seen minimal changes in over 40 years.

For patients who experience disease recurrence or metastasis, outcomes remain poor. Mortality rates following recurrence can reach 80% to 90%, highlighting the urgent need for new treatment options.

These challenges are not unique to osteosarcoma, and many rare cancers face similar obstacles. Limited research funding, small clinical trial populations, and relatively sparse biological data sets are among the most common barriers.

Why Rare Cancers Are Difficult For AI To Address

The same factors that make rare cancers difficult to study also make them challenging for AI systems to analyze. Because these diseases affect small patient populations, the amount of available genetic, clinical, and molecular data is often limited.

AI models trained primarily on large data sets from more common cancers may not be as effective when applied to rare cancer subtypes. If the training data does not include enough examples of a rare cancer, the model may fail to recognize meaningful biological patterns.

In addition to data limitations, economic incentives also shape where AI-driven research is focused. Many AI drug discovery platforms tend to focus on diseases with larger patient populations, where successful therapies can reach more people. While this approach is understandable from a commercial perspective, it can further concentrate innovation around more common cancers.

As a result, rare cancers can become underrepresented in the data sets and systems used to drive modern drug discovery.

How AI Could Better Support Rare Cancer Research

Despite these challenges, AI still has the potential to contribute to rare cancer research if some of these barriers can be addressed.

One important step is expanding access to rare cancer data sets through collaboration. International research networks and partnerships across institutions can help bring together patient data that would otherwise remain spread across individual centers. Even modest increases in available data could significantly improve the ability of these tools to identify relevant patterns.

Another area where progress could be made is in developing AI approaches that work better with smaller data sets. In some cases, these methods allow models to apply what they have learned from larger data sets to situations where less data is available. This may help researchers identify useful patterns in rare cancer data sets that would otherwise be too limited for traditional approaches.

Continued regulatory and policy support also will be important. Programs such as Orphan Drug designation and rare pediatric disease initiatives help encourage investment in rare disease therapies. These programs can help ensure that innovation continues to address patient needs, even when commercial incentives alone may not be enough.

Ensuring Innovation Reaches All Patients

AI will play an increasingly important role in the future of drug discovery. As these tools continue to improve and data sets expand, AI could help accelerate the development of new therapies across many areas of medicine.

At the same time, it’s worth recognizing the limitations of data-driven approaches, especially when it comes to rare diseases. Without better efforts to include rare cancers in research data sets and development strategies, these conditions risk being left out of many advances driven by AI.

Combining advances in AI with continued investment in rare cancer research could help bring new treatment options to patients who currently have very few, while also improving how these diseases are studied and understood over time.

About The Author

Paul Romness, chair, chief executive officer, president of OS Therapies, leads the company with over 25 years of experience in the biopharmaceutical industry, having served every function within major companies like Johnson & Johnson, Amgen, and Boehringer Ingelheim. He has been directly involved in the launch of nine major products in the industry, covering indications for oncology, surgery, HIV, FSD, COPD, IPF, cardiovascular, and diabetes. Throughout his professional career and within his community he has focused on and advocated for unmet medical need and getting treatments to patients. Romness has a B.S. in finance from American University and a Master of Health Policy from George Washington University Medical Center.