People often say that the eyes are the windows of the mind, but Google researchers regard them as indicators of people's health. Google is using deep learning technology to predict a person’s blood pressure, age, and smoking status by analyzing people’s retina images. Google's computer can take clues from the blood vessels, and a previous study showed that computers can use this information to predict whether a person will have a heart attack risk in the near future.
These studies rely on a convolutional neural network, a deep learning algorithm that can change how biologists analyze images. Scientists are using this method to find mutations in genes and predict changes in single cell arrangements. Google has brought a new round of deep learning applications that can make image processing simpler and more versatile, even recognizing previously overlooked biological phenomena.
Philip Nelson, head of engineering at the Google Institute in Mountain View, Calif., said: “The previous application of machine learning to many areas of biology was an unrealistic idea. Now you can do it, and it’s even more exciting that The computer can now observe many details that humans may have never seen."
Convolutional neural networks allow computers to efficiently and completely process images without the need to decompose images. This method first appeared in the technology field in 2012. For example, Facebook used this deep learning technology to identify faces on photos. But scientists have been difficult to apply this method to the field of biology, partly because of the cultural differences between the two fields.
Daphne Koller, chief computer officer at Calico, a San Francisco-based biology company, said: "It's like you're sending a group of biologists to a room where the team of computer scientists is located. They will talk to each other in different languages ​​and they will have different ways of thinking. ."
Scientists must also determine what type of research can be done with convolutional neural networks. When Google wants to use deep learning to find mutations in genes, Google scientists must turn DNA letters into computer-recognizable images. Then they need to train neural networks with reference genes so that mutations can be discovered. The DeepVariant tool introduced in December was able to detect small changes in the DNA sequence. In tests, DeepVariant's performance at least catches up with traditional tools.
Cellular biologists at the Seattle Allen Institute for Cellular Sciences are using convolutional neural networks to convert monotonous gray photos taken with light microscopes into 3D images, and have some organelles with color labels. This method eliminates the process of cell staining. Cell staining is time-consuming and needs to be performed in a precision laboratory, and it will damage the cells. Last month, the team announced a state-of-the-art technology that can predict the shape and location of other parts of the cell using only part of the data.
Anne Carpenter, director of the Broad Institute at MIT and the imaging platform at Harvard University, said: “What you are looking at now is an unprecedented change. Machine learning can accomplish biological tasks with the aid of images.†Her cross in 2015 The subject team began to use the convolutional neural network to process cell images. Now Carpenter claims that about 15% of the image data in her research center is based on convolutional neural networks. She predicts that this method will become the main image processing method of the research center a few years later.
What is even more exciting is that the analysis of images through the use of convolutional neural networks can unwittingly reveal subtle biological phenomena, allowing biologists to begin to think about issues that were previously ignored. According to Rick Horwitz, executive director of the Air Force Institute, such accidental discovery can help medical research advance. If deep learning can uncover the subtle identification of cancer in single cells, it may help researchers identify tumors in advance.
Machine-learning experts in other biology have targeted more advanced fields, and now convolutional neural networks have begun to be widely used in image processing. Alex Wolf, a computer biologist at the German Center for Environmental Health Research, said: "The image is very important, but chemical and molecular data are equally important. I think that in the next few years, a major breakthrough will be made, allowing biologists to use it more widely. Convolutional neural network."
Outdoor Front-service Iron Case Led Wall
Outdoor Front-Service Iron Case Led Wall,Led Display Billboards,Digital Led Display,Advertising Led Wall Media Production
Guangzhou Cheng Wen Photoelectric Technology Co., Ltd. , https://www.cwledwall.com