How Google plans to combat insect-borne diseases with old books

Hundreds of millions of people are affected by insect-borne diseases every year, Google is developing a new technology to combat such diseases using decades-old datasets mined by Google Books. A team at Google Brain is using decades-old datasets mined by Google Books — along with a newly developed sensory map for odour — to combat this global health issue. What made this possible is the recent findings that a mosquito’s sense of smell is not so different from ours.
As former Google Brain researcher, and now Entrepreneur in Residence at Google Ventures, Alex Wiltschko, explains, “My team is focused on giving computers a sense of smell. As we reviewed predictions of the neural networks we trained to predict what molecules smell like to people, we found that they were also useful to predict how the ‘smell parts’ of the brains of insects respond to the same molecules.”
The data challenge
The Google Brain team realised that if they could train computers to recognize the odours that repel mosquitoes, those computers could help predict safe, cheap and effective repellents to stop insect-borne disease in its tracks. However, to train their models (computers), the researchers need data. The team identified relevant research completed by the United States Department of Agriculture during World War II. “We learned about a dataset where they tested thousands of repellents – much more than the 20 we had,” Alex says. However, the USDA data was recorded in notebooks and file cabinets, some of it unindexed, and all of it too difficult to find and share with the Google Brain team. Fortunately, the USDA research had been scanned and indexed by another team at Google – Google Books. The USDA was able to provide search terms which helped the Google Brain team locate the missing dataset among the over 40 million volumes in the Google Books corpus.

Also, when Google Books Technical Collections Specialist Kurt Groetsch heard about that first dataset, he decided to find more. A comprehensive search turned up over 100 datasets of varying sizes. “Scientists did rigorous tests for these research projects. It’s really well-recorded scientific information, and it sits there in print for decades — its existence getting lost in time,” says Google Books Senior Library Partnerships Manager Ben Bunnell. “Now we can take tools that didn’t exist in the ‘40s and use their research to extrapolate information that can potentially save lives today.”
Training the molecules
Armed with all of the data, the Google Brain team tapped TropIQ, an organisation that tests molecules to combat insect-borne infectious diseases, to find out if they could use it. “What we wanted to see was that good repellents in 1942 are still good repellents now — and that’s exactly what we found,” Alex says. With confirmation of the quality of the USDA data, the team was able to train a neural network to predict which molecules were effective as insect repellents. They crossed the molecules currently in use as chemical repellents off the list, then had TropIQ test the remainder: the molecules that had been identified as effective but were not currently in use. Of those, TropIQ’s test revealed 10 molecules that showed a higher level of repellency than DEET. The Google Brain team is currently investigating those molecules for cost, safety and availability, paving the way for a new batch of insect repellents to help combat diseases.