Cancer cells identified with artificial intelligence

Next-generation gene-recognition technology, working much like a staining method, can now identify the molecular features of a cancer cell in a far shorter time. Behind the method lie next-generation technologies such as artificial intelligence and the cloud.

The same kind of progress that smartphone cameras made over the years in low-light photography is now happening in gene technology. In cancer, traditional diagnostic methods were as limited at identifying mutated genes as photographs taken in very low light. Today, work on next-generation gene technologies points to a brighter future in the fight against the disease. In other words, identifying mutated cells is now being addressed with faster and more effective solutions for diagnosis. The work of identification is no longer done in the dark, but in the light.

A cloud-based solution

We spoke about the latest advances in identifying cancer cells with James Creeden, Global Medical Director of Foundation Medicine Inc. (FMI), which Roche acquired for one billion dollars in 2015. Sharing the most recent developments, Creeden explained: "When we look at a patient's comprehensive cancer gene map, we build a sequence covering more than 300 genes. That means generating a great deal of genomic data, and this data is stored partly centrally and partly in cloud-based solutions that comply with the data-privacy regulations in force to protect it. I'd particularly like to stress how much this explosion of data genuinely helps us identify patients who have treatable mutations in their cancers. What's more, these are patients who could not have been diagnosed with the old testing methods."

Just like phone cameras

Asked how the progress paralleling smartphone cameras' low-light performance resembles the new gene-identification process, James Creeden answered: "Absolutely, it's like low-light photography. That's a great example, and one we use often. If you're in a dark room and you've left your keys somewhere and are trying to find them, with a flashlight in your hand you can look everywhere but still may not find them. You need to turn the lights on; you need to see everything as a whole. That's exactly what our technology lets us do across the genome. It turns the light on, and we can see all of the mutations found at once."

Definition, not diagnosis

Underlining that what they do is definition rather than diagnosis, James Creeden said: "For the past 10-20 years, by taking a tumour sample, examining its different layers, looking at it under the microscope and applying what we call basic staining, it has been possible to determine whether it is lung cancer or a particular subtype. In fact, we do not diagnose lung cancer. What our tests do is help define the genomic drivers of the cancer after the doctor has made the diagnosis. This approach can save time and tissue, which are extremely valuable for advanced-stage cancer patients and their families. Personalised, targeted therapies can help the physician in making decisions."

"We were looking for a needle with a flashlight"

Prof. Dr. Mutlu Demiray, Specialist in Internal Medicine and Medical Oncology at Medicana International Istanbul Hospital, whose opinion we also sought, spoke about treatment methods that have turned into personalised care: "The genetics of cancer was always a point of great interest. But searching for a needle in a room with a flashlight was very hard and time-consuming. Once the NGS (next-generation sequencing) technique developed, we were finally able to turn the lights on. With this technique, by gaining insight into the tumour's features from both the tumour tissue and the fragments of tumour DNA shed into the blood, personalised treatments can be designed. We have moved from factory-made treatments to personalised, tailor-made care. An important gain of this technique is the discovery that cancers are not organ-specific — that is, it does not matter whether it is breast, ovary, bowel or lung; it is essentially the changes at the gene level that determine the course of the disease. So we have entered an era of organ-independent molecular definition. In short, if there is a target for smart drugs, which organ the cancer is in no longer matters. Moving to personalised treatments has produced an incredible body of knowledge."

We still have a long way to go

Noting that processing the data and applying the most appropriate treatments to the patient has emerged as a separate challenge, Mutlu Demiray added that cancer types often present with multiple — sometimes 15-20 — mutations at once, and continued: "In this case, processing this data has fallen to machines and powerful computers. In other words, we have reached the era in which machines will learn and guide doctors. We are on the verge of a period in which, once a patient's tumour genetic map is produced, it will be evaluated by computers and presented to the doctor as a preliminary report. Although clear advantages can be obtained for the patient with these methods, we still have a long way to go. We do not yet recommend it routinely for every patient. But for certain cancer types, especially lung cancer, routine recommendations have begun to appear in international guidelines. New genomic data and artificial intelligence will be our most important allies in the fight against cancer."