This week’s list of data news highlights covers March 28-April 3, 2020, and includes articles about using AI to detect misinformation about the 2020 Census and using genetic data to predict outcomes for cancer patients.
Researchers from the University of California, San Francisco, have developed a system that can translate brain activity into speech for 250 words. The researchers trained a neural network to identify patterns of speech, such as vowels, using the brain signal data of four individuals with epilepsy who read sentences aloud. Another network used these patterns to form sentences with 97 percent accuracy.
Researchers from the University of Helsinki have created the COVID-19 Host Genetics Initiative, which is pooling genetic data from around the world to determine which genes affect an individual’s susceptibility to the virus and the severity of their symptoms. The data could help identify drugs that governments could repurpose to treat COVID-19 as well as identify young individuals who are at high risk of developing severe symptoms.
Researchers from New York University and the Chinese hospitals of Wenzhou Central Hospital and Cangnan People’s Hospital have developed an AI-enabled tool that can predict which patients with COVID-19 are likely to develop Acute Respiratory Distress Syndrome, a severe respiratory disease. The researchers developed the tool using patients’ demographic, laboratory, and radiological data, finding that factors such as levels of an enzyme in the liver and hemoglobin were predictive of developing a severe respiratory disease. The tool can predict the risk of developing the respiratory disease with up to 80 percent accuracy.
The U.S. Census Bureau is using an AI-enabled tool to detect social media posts that spread misinformation about the 2020 Census. The tool, which New York software firm Sprinklr developed, allows the agency to respond to and correct misinformation posted online. Sprinklr trained the tool on public social media posts from the past five years.
Locomation, a startup based in Pittsburgh, has created autonomous trucks that transportation company Wilson Logistics will use to move cargo more than 400 miles between Oregon and Idaho this spring. The autonomous vehicles will drive as part of a two-truck convoy, with a human-driven truck that shares data about speed, acceleration, braking, and steering wheel angle leading. The autonomous vehicles also use cameras, LiDAR, and radar to understand their surroundings.
DeepMind, a subsidiary of Google’s parent company Alphabet, has trained an AI agent to play all 57 Atari games. The company trained the agent using reinforcement learning and a rewards system that encourages it to explore before settling on a strategy. In addition, DeepMind gave the agent memory so it could make decisions based on past events. These improvements helped it play games that have long delays between action and reward.
Cordio Medical, an Israeli firm that monitors health conditions by analyzing speech, is partnering with Haifa’s Rambam Hospital in Israel to begin a clinical trial of a tool that can diagnose and monitor COVID-19 patients remotely. The system uses baseline recordings of patients and compares them with later recordings to detect signs of bilateral pneumonia with edema in the lungs, which is associated with COVID-19.
Researchers from the University of Virginia have identified inherited variations in genes that can affect a cancer patient’s response to treatment. The researchers used data from The Cancer Genome Atlas, a large repository of genetic information, and correlated patient outcomes with different variations in genes. The researchers found that identifying a single genetic predictor increased the ability to predict how a patient will respond to therapy by five to ten percent.
Google has created COVID-19 Community Mobility reports, which uses aggregated, anonymous location data to illustrate how people’s movements have changed due to coronavirus. The reports show the percentage increase or decrease of people visiting areas such as parks, grocery stores, and the workplace at the regional and local levels. The reports can help leaders assess the effectiveness of policies that address coronavirus, such as shelter-in-place orders.
Researchers from Kanazawa University in Japan have shown that machine learning can accurately predict which individuals will develop diabetes. The researchers used machine learning to analyze nearly 140,000 patient records, which included data from physical exams, blood and urine tests, and participant questionnaires. The researchers’ model predicted the future development of diabetes with 95 percent accuracy.