NEWS IN FOCUS

differences between fields. “Take a group images of cells captured with light microscopes biological phenomena, prompting biologists to of smart biologists and put them in a room of into 3D images in which some of a cell’s orga- ask questions they might not have considered smart computer scientists and they will talk nelles are labelled in colour. The approach before. “The most interesting phrase in science two different languages to each other, and have eliminates the need to stain cells — a process isn’t ‘Eureka!’, but ‘That’s weird — what’s going different mindsets,” says Daphne Koller, chief that requires more time and a sophisticated lab, on?’” Nelson says. computing officer at Calico — a biotechnology and can damage the cell. Last month, the group Such serendipitous discoveries could help to company in , , that is published details of an advanced technique that advance disease research, says Rick Horwitz, the backed by Google’s parent, Alphabet. can predict the shape and location of even more Allen Institute’s executive director. If deep learn- Scientists also had to identify which types cell parts using just a few pieces of data — such ing can reveal subtle markers of cancer in an of study could be conducted using networks as the cell’s outline (G. R. Johnson et al. Preprint individual cell, he says, it could help to improve that must be trained with huge sets of images at bioRxiv http://doi.org/chwv; 2017). how researchers classify tumour progression. before they can start making predictions. “What you’re seeing now is an unprece- That could in turn trigger new hypotheses about When Google wanted to use deep learning to dented shift in how well machine learning can how cancer spreads. find mutations in genomes, its scientists had to accomplish biological tasks that have to do with Other machine-learning connoisseurs in convert strands of DNA letters into images that imaging,” says Anne Carpenter, director of the biology have set their sights on new frontiers, computers could recognize. Then they trained Imaging Platform at the Broad Institute of MIT now that convolutional neural networks are their network on DNA snippets that had been and Harvard in Cambridge, Massachusetts. In taking flight for image processing. “Imaging aligned with a reference genome, and whose 2015, her interdisciplinary team began to pro- is important, but so is chemistry and molec- mutations were known. The end result was cess cell images using convolutional neural ular data,” says Alex Wolf, a computational DeepVariant, a tool released in December that networks; now, Carpenter says, the networks biologist at the German Research Center for can find small variations in DNA sequences. In process about 15% of image data at her centre. Environmental Health in Neuherberg. Wolf tests, DeepVariant performed at least as well as She predicts that the approach will become the hopes to tweak neural networks so that they conventional tools. centre’s main mode of processing in a few years. can analyse gene expression. “I think there Cell biologists at the Allen Institute for Cell Others are most excited by the idea that will be a very big breakthrough in the next few Science in Seattle, Washington, are using con- analysing images with convolutional neural years,” he says, “that allows biologists to apply volutional neural networks to convert flat, grey networks could inadvertently reveal subtle neural networks much more broadly.” ■

FUNDING billionaire pours funds into high-risk research Silicon Valley philanthropy project revives some grants rejected by US government.

BY EWEN CALLAWAY research. And Chris Somerville, a biochemist Moskovitz, whose estimated net worth is and a scientific adviser to the organization, more than $14 billion, and Tuna have said fter his plan to test a cancer vaccine for says that Open Phil’s total spending will rise that they plan to give away most of their for- middle-aged pet dogs was rejected by several times over the coming years. tune during their lifetimes. It is likely that, the US National Institutes of Health A(NIH), inventor and biochemist Stephen Johnston sought funding outside the main- stream system. On 20 December, the Open Philanthropy Project, a grant-giving organi- zation that is largely funded by Facebook co- JOHN MOORE/GETTY founder Dustin Moskovitz and his wife, Cari Tuna, announced that Johnston will receive US$6.4 million to test the vaccine he devel- oped. His team at Arizona State University in Tempe is now poised to enrol its first pooches in a clinical trial. The science-funding efforts of the Open Philanthropy Project, or Open Phil, have so far flown under the radar compared with those of other Silicon Valley funders. But that is likely to change. The organization, which was launched in 2011 but rebranded under its current name in 2014, has significantly boosted its spending to $200 million this year, of which around $40 million went to scientific Dogs with cancer are about to be enrolled in a clinical trial of a vaccine for the disease.

10 | NATURE | VOL 553 | 4 JANUARY©2 0201818 Mac millan Publishers Li mited, part of Spri nger Nature. All ri ghts reserved. ©2018 Mac millan Publishers Li mited, part of Spri nger Nature. All ri ghts reserved.

IN FOCUS NEWS

in terms of impact on research, Open Phil BIOSECURITY will soon rival better-known philanthropy vehicles, such as the Chan Zuckerberg Initiative in Palo Alto, California, which among other efforts awarded $50 million Ban on pathogen in life-sciences grants in 2017 to create a biohub in the San Francisco Bay Area. Open Phil, based in San Francisco, acknowledges the high odds of failure of studies lifted the basic research it funds and, for a private funder, publishes brutally honest assess- ments of its projects. These range from United States allows work to make viruses more dangerous. developing lab-made meat alternatives to work on a controversial genetic-engineering BY SARA REARDON technology called gene drive. For its latest funding round, Open Phil asked scientists he US government has lifted its contro- whose grant applications had been rejected versial ban on funding experiments that

EYE OF SCIENCE/SPL by an NIH competition for risky research to make certain pathogens more deadly or dust off their proposals. Some 120 research- Ttransmissible. On 19 December, the National ers resubmitted their requests, and it Institutes of Health (NIH) announced that awarded $10.8 million in total to four teams. scientists can once again use federal money to “My hope is Open Philanthropy can conduct ‘gain-of-function’ research on patho- make the world safe for serendipity again,” gens such as influenza viruses. But the agency says Ed Boyden, a neuroscientist at the also said that researchers’ grant applications Massachusetts Institute of Technology in will undergo greater scrutiny than in the past. Cambridge, who won $3 million from the The goal is to standardize “a rigorous pro- project in 2016. He is working to develop cess that we really want to be sure we’re doing a technology that swells tissue to make it right”, says NIH director Francis Collins. Influenza viruses can be modified in the lab. easier to examine under a microscope. The NIH announcement ends a morato- rium on gain-of-function research that began The plan includes a list of suggested factors TAKING A PUNT in October 2014. Back then, some research- for the HHS to consider, including an assess- Gregory Timp, a biophysicist at the ers argued that the agency’s ban — which ment of a project’s risks and benefits, and a University of Notre Dame in South Bend, singled out research on the viruses that cause determination of whether the investigator and Indiana, who has won $2 million from flu, severe acute respiratory syndrome and institution are capable of conducting the work Open Phil to develop a technology to Middle East respiratory syndrome (MERS) safely. It also says that an experiment should sequence proteins, says that the evaluation — was too broad. The 21 projects halted by proceed only if there is no safer alternative process involved rebutting each of the NIH’s the policy included studies of seasonal flu and method of achieving the same results. critiques of his proposal, as well as several efforts to develop vaccines. The NIH eventu- At the end of the assessment process, the rounds of interviews with scientist advisers. ally allowed ten of these studies to proceed, HHS can recommend that the work go ahead, “They have scientific rigour couched in Cal- but three projects using the MERS virus and ask the researchers to modify their plan or sug- ifornia casual. Everything is informal, but eight dealing with flu remained ineligible for gest that the NIH refuse funding. The NIH will they ask these piercing questions,” he says. US government grants — until now. also judge the proposal’s scientific merit before Katherina Rosqueta, founding execu- While the ban was in effect, the NIH and deciding whether to award grant funding. tive director of the Center for High Impact other government agencies examined the costs Scientists have long debated the merits Philanthropy at the University of Pennsyl- and benefits of allowing such research. In 2016, of gain-of-function research and the new vania in Philadelphia, says that the project’s the National Science decision could reopen that discussion. efforts to share its extensive research and Advisory Board for Gain-of- Yoshihiro Kawaoka, a virologist at the Uni- justify its giving makes it stand out among Biosecurity — an function studies versity of Wisconsin–Madison, whose work private funders. “They have a highly independent panel “risked creating was affected by the moratorium, says the analytical view. They have an appetite and that advises the NIH’s an accidental new framework is “an important accomplish- skill in conducting research and sourcing parent agency, the pandemic”. ment”. Kawaoka, who studies how molecular information, and they’re willing to do that US Department of changes in the avian flu virus could make it in a public and transparent way.” Health and Human Services (HHS) — con- easier for birds to pass the infection to humans, Many philanthropists shy away from cluded that very few government-funded gain- now plans to apply for federal funding to basic science because the pay-offs tend to of-function experiments posed a significant experiment with live versions of the virus. be long term and the risks high, says Marc threat to public health. But Marc Lipsitch, an epidemiologist at the Kastner, president of the Science Philan- The new policy outlines a framework that Harvard T.H. Chan School of Public Health thropy Alliance in Palo Alto, a coalition the HHS will use to assess proposed research in Boston, Massachusetts, says that gain-of- of foundations that advocates for private that would create pathogens with pandemic function studies “have done almost nothing funding of basic science. But the Silicon potential. Such work might involve modifying to improve our preparedness for pandem- Valley entrepreneurs who bankroll organi- a virus to infect more species, or recreating a ics — yet they risked creating an accidental zations such as the Open Philanthropy pathogen that has been eradicated in the wild, pandemic”. Project and the Chan Zuckerberg Initiative such as smallpox. There are some exceptions, Lipsitch argues that such experiments are used to long odds, says Kastner. “The however: vaccine development and epidemio­ should not happen at all. But if the government risk-taking is not an issue for them. They logical surveillance do not automatically is going to fund them, he says, it is good that don’t want to be supporting a sure thing.” ■ trigger the HHS review. there will be an extra level of review. ■

©2018 Mac millan Publishers Li mited, part of Spri nger Nature. All ri ghts reserved. ©2018 Mac millan Publishers Li mited, part of Spri nger Nature. All ri ght4s rJANUARYeserved. 2018 | VOL 553 | NATURE | 11