The researchers developed a new computer framework It offers promise in the process of discovering new drugs. Their framework uses an artificial intelligence method called a convolutional neural network to provide global information about new drug candidates.
A team of researchers from Wuhan University published their findings in the journal Big Data Mining and Analytics.
The team developed a fingerprint-embedding framework for drug-target binding affinity prediction (fingerDTA). Calculates fingerprints or profiles of the drug and targets. These targets are molecules that are in some way related to the disease – targets that can be useful in ways that drugs are used to fight a particular disease. The team then used general information from a drug fingerprint or target in a convolutional neural network model and boosted its performance in predicting drug-target binding affinity. FingerDTA is a powerful model for discovering new drugs.
Traditional in vivo Drug discovery, in which researchers work with living organisms to find new drugs to fight diseases, is an expensive and time-consuming process. Researchers can use virtual pre-screening of potential drugs to guide their trials. This virtual process can reduce costs and improve success rates in finding the right medication.
Researchers have widely used two virtual screening methods for drug discovery. One method is high-throughput screening, in which large compound libraries are tested in a short period of time. Another approach involves strategies based on simulated molecular docking, where they study how two or more molecular structures fit together to predict how a protein interacts with small molecules. Although both of these methods have been successfully used in drug discovery, they require extensive experimental design and testing, which is not suitable for large-scale drug screening.
A third method uses drug-target affinity prediction models, where scientists look for a strong attraction between a drug and a target as a means of identifying drugs that may be candidates for treating a disease. This third method has great advantages in efficiency and cost. Scientists have successfully applied deep neural networks to predict drug-target binding relationships. The Wuhan University research team therefore focused their work on a deep learning model for drug-target binding affinity prediction.
Scalability is a key issue when using complex algorithms to analyze large data sets of terabytes or more in a cluster or cloud. A widely used MapReduce type programming model
Often used to process large amounts of data on hundreds or thousands of servers. But MapReduce cannot scale to big data due to memory dependency and high communication cost. The research team proposes a non-map reduce computing framework to improve the scalability of cluster computing on big data. Their framework reduces data communication costs and enables memory-independent approximate computing.
“This new computing framework creates some advantages in big data computing, such as fast sampling of multiple random samples for ensemble machine learning and approximate computing, direct implementation of serial algorithms on local random samples without data communications between nodes, facilitating big data exploration and purification. Also, non-map reduce computing. “Big data simplifies computing and saves energy in cloud computing,” said Professor Juan Liu of Wuhan University’s School of Computer Science.
The research team believes that drug-target binding affinity prediction holds promise in finding new drugs that can prevent viruses from attaching to their targets. “FingerDTA could help find some potential drugs to inactivate COVID-19 by binding to the spike target,” Liu said. It can provide accurate guidance saving considerable manpower and material resources while accelerating new drug research.
Looking ahead, the team hopes to implement the FingerDTA framework on big data platforms and incorporate it into real applications. “Our ultimate goal is to develop such technologies and systems for users to solve application problems of analyzing very large data distributed across multiple data centers,” Liu said.
The research team includes Xuekai Zhu, Liu, Jian Zhang, Zhihui Yang, Feng Yang and Xiaolei Zhang from Wuhan University School of Computer Science.
– This press release was provided by Tsinghua University Press