Cancer is tricky; some would send deceiving messages to our immune system, misleading it for their benefit rather than destruction. Others are nefarious; as they hijack and feed off our brains and nervous systems. In general, cancer is mysterious because it transfigures itself; manifests differently in different bodies and environment. All these turned cancer treatment into some sort of a gamble because both physicians and patients have to bet on the safest or most hopeful resolutions as the malignancy progresses.
Genetic variability affecting treatment results
Technology is perhaps changing the game. Cognitive computing or encouraging machines to have human thoughts had successfully shortened the time needed to uncover new oncology drug candidates. There are past and ongoing researches demonstrating how artificial intelligence (AI) meet or surpass human expert performances lung, skin, and breast cancer screenings, so that precaution or intervention can kick in early. The challenge now lies in establishing a benchmark to ascertain the accuracy of predictions made by non-human.
Recently, researchers from the Francis Crick Institute (Crick), a London based biomedical research center; University of California, Los Angeles (UCLA) Jonsson Comprehensive Cancer Center; Oregon Health & Science University; the Oxford Big Data Institute, and the University of Toronto had co-developed a new open-source software which can dictate the accuracy of predictions made by computers on genetic variation that’s present in tumor samples.
At the moment, there are clinical methods to determine the genetic diversity of tumor samples but they tend to miss out on cell-to-cell variability or genetically identical cells expressing differently in an identical environment. These differences often explain why some patients respond well to a particular treatment while others who share the same condition may not.
Checking on the reliability of algorithms
This new open-source software comes with a simulation framework and scoring system to access the reliability of each algorithm. Specifically, whether these algorithms can accurately forecast, the quantity of cancerous cells; the number of genetically different groups of cancerous cells; the genetic mutations present within each group of cancerous cells, and genetic relations between these groups, within a tumor sample.
580 predictions were created and analyzed as the software was built during the research. The research team had made the software available publicly so that other researchers can use and directly improve upon the framework and scoring system.
Postdoc at the Cancer Genomics Laboratory of Crick and the project’s joint-lead author Maxime Tarabichi said he wished the framework would provide a foundation, whereby overtime as it exposed to more tumor samples and predictions, will become a “much-needed, unbiased, gold-standard benchmarking tool” to assess models that are targeted at signalizing genetic characteristics of tumors.
Likewise, Peter Van Loo, Group Leader at the Cancer Genomics Laboratory of Crick added “computer simulations in cancer genomics are helping us develop more accurate tools, as we understand where these tools perform well, and where they need improvement”. As such, these new technology-driven solutions require more real-life examples, to assist clinicians in providing patients with the best matching or even personalized medicine.