A glioblastoma multiforme (GBM), the most aggressive and fast-growing form of brain tumours, is treated with drugs. To identify which drug has the potential to produce the best results, a genomic study is performed, mapping mutations in the patient's genes. Researchers at IBM Research have succeeded in applying machine learning to this process, allowing them to analyse genome data much faster. They have also experimented with a genome test that examines all of a patient's genes.
In Neurology Genetics, a magazine of the American Academy of Neurology, the researchers describe their findings. GBMs are the most aggressive form of gliomas, tumours in the brain that arise from glia cells. These are the support cells that nourish, protect and insulate nerve tissue. IBM Research reports that hundreds of thousands of brain tumours are detected each year in the United States, a quarter of which are gliomas. Almost 50% of all gliomas are GBMs. GBMs grow very aggressively and, in many cases, invade surrounding tissue.
Genome testing
To treat GBMs, genome tests are performed, which map mutations in genes. Using a database, doctors can look for a suitable drug in these mutations. However, many commercial genome tests only look at a small proportion of a patient's genes. This since accurately assessing genome data is very time-consuming. Researchers at IBM Research have investigated whether analysing more genes gives a more complete picture of the patient and could yield gains in the treatment of GBMs.
The researchers investigated this by developing a genome test that tests all of a patient's genes. Experiments with this test show that such a more comprehensive test can identify more mutations in genes, giving oncologists more potential drugs to treat GBMs with. Which treatment is most suitable for the specific patient, however, still has to be determined manually. This includes looking at results previously achieved with these specific drugs in treating GBMs or other cancers, and whether the drugs have been approved by, for example, the Medicines Evaluation Board (CBG) for their treatment.
Very time-consuming
Analysing a genome test that examines all genes takes as many as 160 man hours on average. This makes such tests too time-consuming to deploy on a large scale. IBM Research researchers have therefore unleashed IBM Watson for Genomics on genome data. This is an analytical tool from IBM specifically trained in analysing genome data. IBM Watson is a supercomputer that uses machine learning to gain knowledge. Researchers let IBM Watson process millions of documents, such as encyclopaedias, research reports and other reference material. Based on these documents, IBM Watson can improve its knowledge. The more data the system is presented with, the more accurate it can do its job.
The deployment of IBM Watson for Genomics is proving to save considerable time. Whereas human specialists need 160 man hours to assess the data, IBM Watson for Genomics can perform this process in just 10 minutes. This without compromising on accuracy. The researchers at IBM Research argue that this time saving is crucial for conducting such comprehensive genome testing on a large scale. The researchers therefore call their research an important step forward in promoting human-machine interface as a way to address a major bottleneck in cancer genomics.
Author: Wouter Hoeffnagel
Source: IBM Research
Source: Neurology Genetics
