THE BODY MECHANIC:
Jacqueline Detwiler on whether all that individualised cancer data can really help patients.
CASE STUDY 1 The sharing project
THREE YEARS AGO, Dr Jinghui Zhang, chair of the department of computational biology g at St Jude Children’s Research Hospital in i Memphis, tried to download a set of 1 547 samples from a government database so she could run an experiment on it. The download took a year and a half. Every time Zhang’s team thought it had the complete set of data, they then performed a statistical check that was supposed to catch any transmission errors. Seven times, the check failed. Seven times, the team started working only to realise that part of the set was missing, and they’d have to start all over. ‘Data downloading is a very painful process,’ Zhang says. ‘It can potentially drive researchers away from science. Period.’
Zhang had already started to wonder how much faster cancer research could move if data sets were freely and immediately available to the computational biologists who wanted to work on them, so in April she partnered with Microsoft to release St Jude Cloud, a web-based data processor that offers access – within 48 hours – to the largest publicly available set of childhood cancer stats in the world, the genomes of more than 5 000 St Jude patients. The sharing is so generous that it defies academic logic. ‘People think that there are strings attached to this,’ says Zhang. ‘They just don’t believe that we would do such a thing.’
CASE STUDY 2 Machine learning for doctors
ST JUDE is also not the only organisation that has seized on sharing and collaboration as a solution to stalled cancer research. In January 2018, Microsoft also partnered with the charity Stand Up to Cancer on an $11 million (R160 million) research program called Convergence 2.0 to solve a problem that is essentially the reverse of the one Zhang faced: researchers who have data sets or ideas worth studying, but little or no computer-programming expertise. ‘There was hardly a cancer centre in the world that had the same people that Facebook, Microsoft, Amazon, and Google were getting to do deep-learning algorithms. These are two nonoverlapping groups of people,’ says Arnie Levine, professor emeritus at the Institute for Advanced Study and co-vice chairperson of Stand Up to Cancer’s scientific advisory committee. ‘So what we did was we made the marriage.’ Convergence 2.0 funded seven teams of researchers and medical doctors, matching each with machine-learning experts from places like Microsoft and MIT.
3 CASE STUDY Sharable tumour samples
MEANWHILE, for clinicians, , a new company called Paige.ai Paige.gai made an agreement g with Memorial Sloan Kettering Cancer Center in i New York to train machine-learning algorithms on its absolutely massive collection of 25 million fully digitised tumour slides. The software will eventually be able to help doctors all over the country diagnose cancer from biopsies the same way a top MSK pathologist would, only with even more accuracy, objectivity, and information about treatment options and potential survival.
UPDATE Jacqueline Detwiler also reported on the state of cancer research in our September 2018 issue.