Home BlogWeekly News 10 Bits: the Data News Hotlist

10 Bits: the Data News Hotlist

by Cassidy Chansirik
Bus in Washington D.C.

This week’s list of data news highlights covers November 28, 2020 – December 4, 2020 and includes articles about predicting protein shapes with AI and improving public transit operations with machine learning.

1. Predicting Protein Shapes with AI

Researchers from DeepMind, a Google-owned AI research laboratory, have developed an AI system that uses neural networks to predict a protein’s 3D shape based on its amino acid sequence. By determining the 3D shapes of proteins stands, Google’s deep learning program has made a significant leap in solving one of biology’s grandest challenges, allowing researchers to understand the building blocks of cells to accelerate more advanced drug discovery. Testing showed that 67 percent of the system’s predictions matched lab experiments, with a margin of error comparable to the width of an atom.

2. Identifying Gaps in Vaccine Coverage with Machine Learning

Researchers from the Massachusetts Institute of Technology have used machine learning models to identify disparities in how effective coronavirus vaccines will be for people from different ethnic groups. The researchers trained the models to predict whether a vaccine could stimulate an immune response in people with different ethnicities using patient data and models of proteins in their immune systems. They concluded that vaccines from Moderna, Pfizer, and AstraZeneca may be less effective for people of Black or Asian ancestry because these vaccines were not developed with a diverse set of viral particles.

3. Developing a Balloon Navigation System Powered by AI

Loon, a subsidiary of Alphabet that focuses on providing Internet access to rural areas using high-altitude balloons in the stratosphere, are using AI to steer balloons more accurately and efficiently. Developers trained the navigation system using reinforcement learning, a machine learning technique that uses trial and error in order to improve its decision making abilities. In a test conducted over 39 days, the system used less power to keep a balloon over a defined location and learned how to navigate to a defined location more accurately than a human designed navigation system. 

4. Identifying Patients with Alzheimer’s Disease with Machine Learning

Researchers from Duke University in North Carolina have developed a machine learning tool that uses retinal images to identify Alzheimer’s disease in symptomatic individuals. Because Alzheimer’s can affect an individual’s vision, depth perception, and spatial relationships, the researchers trained the tool to identify changes in blood vessel density based on 636 retinal image scans, 144 of which were from patients already diagnosed with Alzheimer’s. The tool achieved the highest accuracy when it was combined with patient information and is providing clinicians with a predictive tool that is non-invasive and more cost effective than existing diagnostic tests. 

5. Improving Public Transit Operations with Machine Learning

Chattanooga Area Regional Transit Authority (CARTA) has partnered with Vanderbilt University in Tennessee to improve public transit operations with machine learning models. CARTA is using machine learning to identify which diesel bus routes can be replaced by battery-electric buses. The models take into account variables such as climate, the distance of a route, the inclination of a road, and how much it will cost to charge the batteries the buses will use. By introducing battery-electric busses, CARTA hopes to increase operational efficiency and increase public ridership. 

6. Building an AI Algorithm That Designs Its Own Robot Bodies

Engineers at the Massachusetts Institute of Technology have developed RoboGrammer, an AI algorithm that designs robots optimized to traverse difficult terrains, such as a slippery floor or a set of stairs. The system works by designing robots inspired by the bodies of animals. For example, the algorithm built a robot with a lizard-like body for smooth terrain and designed a robot that can pull itself forward with two arms like a walrus for icy surfaces. 

7. Developing Cancer Therapies with Machine Learning

Researchers from the University of California San Francisco and Princeton University are using machine learning to develop cancer therapies. The machine learning models examined the activity and genetic markers of 2,300 genes found in normal and tumorous cells to identify the protein combinations that are activated in cancerous cells. With this information, researchers can develop cancer cell therapies to precisely target tumor cells while leaving surrounding healthy tissues undamaged. 

8. Transforming Traditional Surveillance Cameras in Businesses with Machine Learning

Amazon has developed Panorama, a hardware device that uses machine learning and computer vision to help businesses better monitor their business operations. With Panorama, businesses can run computer vision models on cameras to automate a variety of tasks such as inspecting parts on a manufacturing line, ensuring employees are following safety protocols, such as wearing hardhats or protective goggles, or analyzing traffic in retail stores. 

9. Assisting College-Bound Students with AI

The Texas Higher Education Coordinating Board has partnered with Educate Texas, a nonprofit focused on community education initiatives, to develop ADVi, an AI chatbot that answers questions about college applications and provides information on the Free Application for Federal Student Aid (FAFSA), which U.S. students complete to receive federal and state financial aid while attending college. Given Texas received 18 percent fewer FAFSA applications this year, the need for college application support resources has grown. Students can interact with ADVi by sending questions via text, such as asking when FAFSA is due, and ADVi responds by directing students to appropriate websites

10. Identifying Allergic Reactions in Patients with Deep Learning

Researchers at the Brigham and Women’s Hospital in Massachusetts have developed a deep learning algorithm to identify allergic reactions in the free-text narrative of hospital patient safety reports. Researchers trained the algorithm based on 101 keywords clinicians at Massachusetts General Hospital commonly use to identify allergy symptoms to medications, foods, or healthcare products. When tested, the algorithm reduced the number of cases for manual review by 63 percent and identified 24 percent more confirmed cases of allergic reactions compared to clinicians.

 

Image credit: Mario Sessions

You may also like

Show Buttons
Hide Buttons