Skip to main content

IBM R&D Labs in Israel News

IBM Research and EU consortium make online system available to help physicians choose optimal HIV treatment

IBM, along with other partners in the EuResist consortium, are announcing the availability of a European integrated system for clinical management of antiretroviral drug resistance. The online system, now open to the global medical community, provides clinicians with a prediction of response to antiretroviral treatment for HIV patients. The project’s new technologies and mathematical models are providing a smarter and more efficient way to choose the best drugs and drug combinations for any given HIV genetic variant.

Although there have been tremendous leaps and bounds in the management of the virus that causes AIDS, HIV has the ability to develop resistance to any antiretroviral compound. This means that doctors need to continuously monitor and prescribe new therapies for the treatment to remain effective. Keeping HIV under control is dependant on choosing the right combination of drugs that work for the longest period of time. Importantly, there is solid evidence that patients harboring multidrug resistant virus are at increased risk of AIDS and death.

Researchers behind the EU-funded EuResist project have developed new mathematical prediction models that not only take into account the patient’s own history, but tap into the wealth of information that EuResist researchers have amassed. The recent expansion of the EuResist database to include information from more than 33,000 patients and 98,000 therapies, and 370,000 viral load measurements—makes it the world’s biggest database centered on HIV resistance and clinical response information.

“By combining some of the largest databases and creating new prediction engines, the project can provide a prediction of how HIV will react in a certain person given a certain combination of drugs,” explained Francesca Incardona, EuResist’s coordinator. “This system outperforms the current state-of-the-art- predictive system available to medical researchers.”

The researchers evaluated the EuResist prediction engines separately and in combination, and found that the EuResist model was 76 percent accurate, outperforming other commonly used HIV resistance prediction databases.

Even more interestingly, the system also outperformed expert humans. The EVE (Engine versus Experts) study compared EuResist with ten international experts confronted with 25 case histories where all the clinical and virological information was available, an attempt to simulate real practice in HIV specialized care. EuResist and only one of the ten experts made six incorrect predictions while all the other experts had more errors.

The system is currently available online through a web interface that is freely available to the global medical community at

“The achievement of EuResist will help bring about better medicine, lower treatment-related toxicity and cost savings – which means higher quality of care for the millions of people worldwide who are infected with the virus,” noted Dr. Michal Rosen-Zvi, IBM machine learning researcher for the EuResist project.

In fact, the ICT ( Information and Communication Technologies) Health unit of the European Commission recently took a detailed look at the success garnered by the project and selected EuResist as the ‘Project of the Month’ for November 2008.

How It Works

The project’s biomedical information integration technology gathers data from three large genotype-response databases, namely the Italian ARCA database (one of the biggest in the world), the German AREVIR database, and data coming from the Karolinska Infectious Diseases and Clinical Virology department. Recent expansion includes also data from Luxembourg, Belgium (Leuven) and Spain (Catalonia). The data include treatment histories, treatment response information, and the sequence of the relevant part of the HIV genome (genotype).

EuResist addresses not only the genotype, which can predict what happens with the virus, but also uses other relevant information from the patient history. The technology checks which drugs were already administered to the patient in the past and analyzes how previous treatments affect the success of the predicted treatment. Information is retrieved from the EuResist database for past treatments and correlated with the success or failure of the recommended treatment.

IBM researchers at the Haifa Lab are contributing to two aspects of the EuResist project. The Healthcare and Life Science group has implemented a standardized biomedical information technology that processes and correlates clinical and genomic data from the various data sources. The Machine Learning group worked on the sophisticated model and training engine that helps predict drug resistance.


Aside from the IBM Research Lab in Haifa, the other partners participating in this European Union 6th Framework project include: Informa S.r.l., Università degli Studi di Siena, Italy; Karolinska Institute, Sweden; Max Planck Institute for Informatics, Saarbrücken, germany, University Hospital of Cologne, Germany; RMKI, Hungary; Kingston University, UK; and the European Federation of Pharmaceutical Industries and Associations (EFPIA).