Amazing Science

Huge crowdsourced genealogy databases are inspiring new genetics research.

Yaniv Erlich has been a white-hat hacker and a geneticist at Columbia University, and now he works for a genealogy company.

This unusual career trajectory has led, most recently, to a 13-million-person family tree unveiled recently in Science. The massive trove of data comes from public profiles on the crowdsourced genealogy website Geni.com, and it sheds light on human longevity and dispersal over time. (I wrote about a preprint of this paper last year.) But most of all, Erlich is excited about overlaying DNA information on top of family trees to study genes implicated in disease.

MyHeritage, the company behind Geni.com, also sells DNA ancestry tests. And since 2017, Erlich has been on leave from Columbia working as MyHeritage’s chief scientific officer to develop those DNA tests.

If that sounds like a lot of data going into the hands of one company, well, it is. Erlich has very much been in discussions about DNA research and privacy. In 2013, he showed it was possible, using only public information in places like consumer genealogy databases, to identify certain study participants who had contributed their DNA to research projects. For this feat, Nature dubbed him the “genome hacker.”

It’s hard to pinpoint the dawn of AI gaming supremacy. You could choose Kasparov’s loss in chess, or Lee Sedol’s defeat at the virtual hands of AlphaGo. Another popular option would be when legendaryJeopardy! champion Ken Jennings lost to IBM’s Watson in 2011. Watson could parse the game’s clues and handle wordplay. The two-day match wasn’t close. “I for one welcome our new computer overlords,” Jennings wrote under his final answer.

Watson seemed to be endowed with the kind of clerical skills humans use on a host of real-world problems. It could take a prompt in English, rummage through relevant documents at lightning speed, come up with the relevant snippets of information, and settle on a single best answer. But seven years later, the real world continues to present stubborn challenges for AI. A September report by the health publication Stat found that researching and designing personalized cancer treatments, as Watson’s descendant Watson for Oncology aims to do, is proving difficult.

“The questions in Jeopardy! are easier in the sense that they don’t need much common sense,” wrote Bengio, who has collaborated with the Watson team, when asked to compare the two cases from the AI perspective. “Understanding a medical article is much harder. Again, much basic research is needed.”

As special as games are, there are still a few real-world problems they resemble. Researchers from DeepMind declined to be interviewed for this article, citing the fact that their AlphaZero work is currently under peer review. But the team has suggested that its techniques may soon help biomedical researchers, who would like to understand protein folding.

To do this, they need to figure out how the various amino acids that make up a protein kink and fold into a little three-dimensional machine with a function that depends on its shape. That’s tricky in the same ways chess is tricky: Chemists know the rules roughly well enough to calculate specific scenarios, but there are still so many possible configurations, it’s a hopeless task to search through them all. But what if protein folding could be configured as a game? In fact, it already has been. Since 2008, hundreds of thousands of human players have attempted Foldit, an online game where users are scored on the stability and feasibility of the protein structures they fold. A machine could train itself in a similar manner, perhaps by trying to beat its previous best score with general reinforcement learning.

Deep-learning researchers have found that certain neurons in the brain have shape and electrical properties that appear to be well-suited for “deep learning” — the kind of machine-intelligence used in beating humans at Go and Chess.

Canadian Institute For Advanced Research (CIFAR) Fellow Blake Richards and his colleagues — Jordan Guerguiev at the University of Toronto, Scarborough, and Timothy Lillicrap at Google DeepMind — developed an algorithm that simulates how a deep-learning network could work in our brains. It represents a biologically realistic way by which real brains could do deep learning.

The finding is detailed in a study published December 5th in the open-access journal eLife. (The paper is highly technical; Adam Shai of Stanford University and Matthew E. Larkum of Humboldt University, Germany wrote a more accessible paper summarizing the ideas, published in the same eLife issue.)

Analyses across dozens of cancers reveal new potential drug targets.

Researchers have combed through the genomes of more than 1,700 tumors, representing more than 20 different kinds of childhood cancer, to unearth potential drug targets and a better understanding of how cancer arises in the youngest patients.

The work, published by two teams on 28 February in Nature1,2, is part of a growing movement to harness genomics against childhood cancers. Around the world, researchers are banding together to share genomics data, techniques and cell lines that can be used to study these diseases.

The efforts are generating a long list of genetic alterations that might be important in triggering cancer. But it will take considerable work to determine which of these genetic clues are pointing to the best drug targets. “It’s not just a matter of picking through genomics data,” says Susan Weiner, founder of the Children’s Cause for Cancer Advocacy in Washington DC. “There’s also the question of the validity of the targets.”

Microbes that produce methane may already be living on Enceladus, a moon of Saturn which is tipped to host life because it boasts a liquid water ocean beneath a crust of solid ice, and strange atmospheric plumes of water. That’s the implication of research showing that an earthbound organism which also produces methane can happily survive in conditions known to exist on Enceladus, from observations by the Cassini space probe before its mission ended last year.

Isolated from deep sea vents almost 1000 metres deep in the Okinawa Trough off Japan, Methanothermococcus okinawensis was subjected to gruelling physical and chemical conditions found on Enceladus for more than five years.

This microbe, called a methanogenic archaeon, survives without oxygen by combining hydrogen and carbon dioxide – both observed in Enceladus’s atmosphere – to make the energy it needs, emitting methane as a waste product. Cassini detected traces of methane in Enceladus’s plumes, and there’s a chance that some of it may have come from this kind of microbe.

“The conditions we mimicked in the lab are as close as possible to those inferred from Cassini on Enceladus,” saysSimon Rittmann at the University of Vienna in Austria, who led the investigations.

Rittmann subjected the microbe to various combinations of gases found on Enceladus, and found that it was always able to survive when provided with the moon’s levels of hydrogen and carbon dioxide. It was still able to thrive at temperatures and pressures likely found in Enceladus’s oceans, ranging respectively from 0 to 90 degrees Celsius, and up to 50 Earth atmospheres.

New finding suggests differences in how humans and bacteria control production of DNA’s building blocks.

Using a state-of-the-art type of electron microscopy, an MIT-led team has discovered the structure of an enzyme that is crucial for maintaining an adequate supply of DNA building blocks in human cells. Their new structure also reveals the likely mechanism for how cells regulate the enzyme, known as ribonucleotide reductase (RNR). Significantly, the mechanism appears to differ from that of the bacterial version of the enzyme, suggesting that it could be possible to design antibiotics that selectively block the bacterial enzyme.

“People have been trying to figure out whether there is something different enough that you could inhibit bacterial enzymes and not the human version,” says Catherine Drennan, an MIT professor of chemistry and biology and a Howard Hughes Medical Institute Investigator. “By considering these key enzymes and figuring out what are the differences and similarities, we can see if there’s anything in the bacterial enzyme that could be targeted with small-molecule drugs.”

Drennan is one of the senior authors of the study, which appears in the Feb. 20 issue of the journal eLife. JoAnne Stubbe, the Novartis Professor of Chemistry Emerita at MIT, and Francisco Asturias, an associate professor of biochemistry at the University of Colorado School of Medicine, are also senior authors. The paper’s lead authors are MIT research scientist Edward Brignole and former Scripps Research Institute postdoc Kuang-Lei Tsai, who is now an assistant professor at the University of Texas Houston Medical Center.

The RNR enzyme, which is found in all living cells, converts ribonucleotides (the building blocks of RNA) to deoxyribonucleotides (the building blocks of DNA). Cells must keep a sufficient stockpile of these building blocks, but when they accumulate too many, RNR is shut off by a deoxynucleotide molecule known as dATP. When more deoxynucleotides are needed, a related molecule called ATP binds to RNR and turns it back on.

An unusual feature of RNR is that it can catalyze the production of four different products: the nucleotide bases often abbreviated as A, G, C, and T. In 2016, Drennan discovered that the enzyme achieves this by changing its shape in response to regulatory molecules. Most of the researchers’ previous work on RNR structure has focused on the version found inE. coli. For those studies, they used X-ray crystallography, a technique that can reveal the atomic and molecular structure of a protein after it has been crystallized.

Targeted motor and sensory reinnervation (TMSR) is a surgical procedure on patients with amputations that reroutes residual limb nerves towards intact muscles and skin in order to fit them with a limb prosthesis allowing unprecedented control. By its nature, TMSR changes the way the brain processes motor control and somatosensory input; however the detailed brain mechanisms have never been investigated before and the success of TMSR prostheses will depend on our ability to understand the ways the brain re-maps these pathways. Now, EPFL scientists have used ultra-high field 7 Tesla fMRI to show how TMSR affects upper-limb representations in the brains of patients with amputations, in particular in primary motor cortex and the somatosensory cortex and regions processing more complex brain functions. The findings are published in Brain.

Targeted muscle and sensory reinnervation (TMSR) is used to improve the control of upper limb prostheses. Residual nerves from the amputated limb are transferred to reinnervate and activate new muscle targets. This way, a patient fitted with a TMSR prosthetic "sends" motor commands to the re-innervated muscles, where his or her movement intentions are decoded and sent to the prosthetic limb. On the other hand, direct stimulation of the skin over the re-innervated muscles is sent back to the brain, inducing touch perception on the missing limb.

But how does the brain encode and integrate such artificial touch and movements of the prosthetic limb? How does this impact our ability to better integrate and control prosthetics? Achieving and fine-tuning such control depends on knowing how the patient's brain re-maps various motor and somatosensory pathways in the motor cortex and the somatosensory cortex.

A major trend at NIPS this year is the rise of unsupervised learning and generative modeling. A groundbreaking example of this is the paper “Unsupervised Image-to-Image Translation Networks,” led by NVIDIA researcher Ming-Yu Liu. To date, much of deep learning has used supervised learning to provide machines a human-like object recognition capability. For example, supervised learning can do a good job telling the difference between a Corgi and a German Shepherd, and labeled images of both breeds are readily available for training.

To give machines a more “imaginative” capability, such as imagining how a wintery scene would look like in the summer, Liu and team used unsupervised learning and generative modeling. An example of their work is shown below, where the winter and sunny scenes on the left are the inputs and the imagined corresponding summer and rainy scenes are on the right. The NVIDIA Research team’s work uses a pair of generative adversarial networks (GANs) with a shared latent space assumption to obtain these stunning results. Considering the top two images above, the first GAN is trained on the winter scene — overcast skies, bare trees, snow covering just about everything but the cars sailing down the frozen road. The second GAN is trained to understand generally what summer looks like, but hasn’t been trained on the same specific scene as its counterpart.

And how could it? You’d need the same footage recorded from the same vantage point, with the same perspective and with all the oncoming traffic and other details in exact same location — for both summer and winter. The unsupervised learning developed by the team removes the need for this capture and labeling, which would otherwise take extensive time and manpower.

This unsupervised translation is enabled through a shared latent space assumption, which associates the GANs with each other by tying some of their parameters together. A summery translation of the wintery scene can be generated by transferring the representation from the first GAN to the second.

The use of GANs isn’t novel in unsupervised learning, but the NVIDIA research produced results — with shadows peeking through thick foliage under partly cloudy skies — far ahead of anything seen before. The potential benefits of this technique are widespread. In addition to needing less labeled data and the associated time and effort to create and ingest it, deep learning experts can apply the technique across domains. For self-driving cars alone, training data could be captured once and then simulated across a variety of virtual conditions: sunny, cloudy, snowy, rainy, nighttime, etc.

A team of scientists has deployed a form of DNA to create a circuit element capable of splitting and combining current—much like an adapter that can connect multiple appliances to a wall outlet.

In a new study, a special form, known as G-quadruplex (G4) DNA, is used to improve charge transport properties. G4 DNA is composed of four rather than two strands of DNA that contain only the base guanine. “DNA is capable of conducting charge, but to be useful for nanoelectronics, it must be able to direct charge along more than one path by splitting or combining it,” says Peng Zhang, an assistant research professor of chemistry at Duke University and a co-author of the study. “We have solved this problem by using the guanine quadruplex (G4) in which a charge can arrive on a duplex on one side of this unit and go out either of two duplexes on the other side.”

“Various device-like functions have been reported for two-terminal single molecular junctions, but a practical circuit-like element requires three or more terminals for control and for electrical networking,” notes Seeman. “This work demonstrates that a strategically designed G-quadruplex motif can fulfill the critical requirements for moving beyond two-terminal functionality.”

“This is the first step needed to transport charge through a branching structure made exclusively of DNA,” Zhang adds. “It is likely that further steps will result in successful DNA-based nanoelectronics that include transistor-like devices in self-assembling ‘pre-programmed’ materials.”

Technology that mimics the functions and sensations of human skin may sound futuristic, but it isn’t exactly new. In 2010, scientists at Stanford University and at University of California, Berkeley, developed artificial skin that could feel even a gentle touch. Dubbed e-skin, the device holds great potential to help future robots more delicately interact with humans. 

But now, researchers at the University of Colorado Boulder are giving e-skin an environmentally friendly upgrade. As Adam Epstein writes for Quartz, the latest iteration of the device can re-heal itself and be recycled to make new layers of skin. The recyclable e-skin, which was described last week in the journal Science Advances, looks like a thin, yellow adhesive sticker. Embedded in this thin layer are sensors that measure pressure, temperature, humidity and air flow. It’s made out of a polymer and silver nanoparticles, which allows it to bend and move with curved spaces like fingers or toes. It’s also soft, but not quite as stretchy as human skin, The Verge’s Alessandra Potenza reports. 

Many labs around the world are creating e-skin. But what makes this e-skin special, Potenza points out, is that it’s virtually waste-free. It heals itself through a chemical reaction between the compounds that make up the bandage-like film. If it is sliced, the researchers can simply just add more of the original three active compounds that make up the film and the ensuing chemical reaction will knit the e-skin back together, “healing” the wound, Potenza reports. If ever damaged beyond repair, it can also be soaked in a solution that separates out the silver nanoparticles and then be recycled into new e-skin. 

The healing process happens fast, too. It only takes a few minutes at 140 degrees Fahrenheit or 30 min at room temperature, while the recycling process takes about 30 minutes at 140 degrees or 10 hours at room temperature.

Edinburgh-based Skyrora, a company with partners in Ukraine, will launch a partially 3D printed suborbital vehicle from the north of Scotland later this year. Skyrora’s rocket engines run on hydrogen peroxide and kerosene.

America, Russia, China, and… Scotland? Yes, the newest contender in the space race is that tiny northern country of the United Kingdom, Scotland, whose very own Skyrora has developed a suborbital launch vehicle that will take off from northern Scotland in the last three months of 2018.

It’s a great achievement for the company, which also operates from Ukraine and which has used 3D printing to develop parts for its spacecraft. Potential launch locations include Shetland, where the Shetland Space Centre is bidding for a license from the UK Space Agency. If the company can secure a launch location, its partly 3D printed suborbital launch vehicle could be taking off within the year.

University of Washington and Microsoft researchers revealed today that they have taken a significant step forward in their quest to develop a DNA-based storage system for digital data. In a paper published in Nature Biotechnology, the members of the Molecular Information Systems Laboratory (MISL) describe the science behind their world record-setting achievement of 200 megabytes stored in synthetic DNA. They also present their system for random access — that is, the selective retrieval of individual data files encoded in more than 13 million DNA oligonucleotides. While this is not the first time researchers have achieved random access in DNA, the UW and Microsoft team have produced the first demonstration of random access at such a large scale.

One of the big advantages to DNA as a digital storage medium is its ability to store vast quantities of information, with a raw limit of one exabyte — equivalent to one billion gigabytes — per cubic millimeter. The data must be converted from digital 0s and 1s to the molecules of DNA: adenine, thymine, cytosine, and guanine. To restore the data to its digital form, the DNA is sequenced and the files decoded back to 0s and 1s. This process becomes more daunting as the amount of data increases — without the ability to perform random access, the entire dataset would have to be sequenced and decoded in bulk in order to find and retrieve specific files. In addition, the DNA synthesis and sequencing processes are error-prone, which can result in data loss.

MISL researchers addressed these problems by designing and validating an extensive library of primers for use in conjunction with polymerase chain reaction (PCR) to achieve random access. Before synthesizing the DNA containing data from a file, the researchers appended both ends of each DNA sequence with PCR primer targets from the primer library. They then used these primers later to select the desired strands through random access, and used a new algorithm designed to more efficiently decode and restore the data to its original, digital state.

“Our work reduces the effort, both in sequencing capacity and in processing, to completely recover information stored in DNA,” explained Microsoft Senior Researcher Sergey Yekhanin, who was instrumental in creating the codec and algorithms used to achieve the team’s results. “For the latter, we have devised new algorithms that are more tolerant to errors in writing and reading DNA sequences to minimize the effort in recovering this information.”

Using synthetic DNA supplied by Twist Bioscience, the MISL team encoded and successfully retrieved 35 distinct files ranging in size from 29 kilobytes to over 44 megabytes — amounting to a record-setting 200 megabytes of high-definition video, audio, images, and text. This represents a significant increase over the previous record of 22 megabytes set by researchers from Harvard Medical School and Technicolor Research & Innovation in Germany.

“The intersection of biotech and computer architecture is incredibly promising and we are excited to detail our results to the community,” said Allen School professor Luis Ceze, who co-leads the MISL. “Since this paper was submitted for publication we have reached over 400 megabytes, and we are still growing and learning more about large-scale DNA data storage.”