Molecular screening efficiency reaches twice the industry standard. New machine learning algorithms help speed drug development.
February 14, 2019 Source: Science and Technology Daily
Window._bd_share_config={ "common":{ "bdSnsKey":{ },"bdText":"","bdMini":"2","bdMiniList":false,"bdPic":"","bdStyle":" 0","bdSize":"16"},"share":{ }};with(document)0[(getElementsByTagName('head')[0]||body).appendChild(createElement('script')) .src='http://bdimg.share.baidu.com/static/api/js/share.js?v=89860593.js?cdnversion='+~(-new Date()/36e5)];Researchers at the University of Cambridge in the UK have designed a new machine learning algorithm to find drugs that have proven to be twice as efficient as current industry standards, helping to speed up the development of new drugs. The research results were published in the recently published Proceedings of the National Academy of Sciences.
The key point of drug discovery is to predict whether a molecule will activate a particular physiological process. Statistical models can be established by searching for chemical patterns shared between molecules that activate physiological processes, but the data currently building these models is very limited because of the high cost of the experiment and the unclear which chemical patterns are statistically significant. “Machine learning has made significant progress in areas such as data-rich computer vision.†The main person in charge of the project, Dr. Alpha Li of the Cavendish Laboratory at Cambridge University, said that it is used in the field of drug discovery. Solve the problem of relatively limited data volume.
Known as the mathematical principle of random matrix theory, it gives assumptions about the statistical properties of random and noisy data sets. Using this principle, statistical data on the chemical characteristics of active and inactive molecules can be compared to determine which chemical modes are combined. In terms of what is really important, what is only accidental.
Based on this assumption, the research team worked with Pfizer to develop an algorithm that uses mathematical operations to separate pharmacologically related chemical patterns from unrelated chemical patterns. Importantly, the algorithm not only studies molecules that are known to be active, but also does not let those inactive molecules, and learns which parts of the molecule are important for drug action, and which parts are not important, making those Failed experiments (data) can also be used effectively.
The researchers began modeling with 222 active molecules and were able to screen the other 6 million molecules from a computational perspective. As a result, the researchers screened out 100 of the most relevant molecules, and found four new molecules that activate the CHRM1 receptor that may be associated with Alzheimer's disease and schizophrenia.
“Screening 4 active molecules from 6 million molecules is like finding a needle in a haystack,†Dr. Li said. “The detailed comparison shows that the efficiency of the new algorithm is twice that of the industry standard.†Researchers are currently refining the algorithm to predict Methods for synthesizing complex organic molecules and extending machine learning methods to new material design fields. (Reporter Tian)
Chest Seal,Chest Seal Kit,Chest Seal First Aid,Chest Seals Medical
Roosin Medical Co.,Ltd , https://www.roosinmedical.com