Researchers at Clemson University have developed a deep learning tool that provides greater insight into the influence of gene regulatory networks (GRNs) on drug response. The researchers recently published the results in the journal Nature Biotechnology.
GRNs are complex systems of interactions between genes, regulatory elements and proteins. They can help us understand how genetic differences affect traits such as the way people respond to drugs,” Zhana Duren, PhD, assistant professor in the Department of Genetics and Biochemistry at Clemson University, explained in a press release. “Every person has a unique GRN that is formed by the specific genotype, or set of genetic characteristics, of that person. That explains why people can react differently to the same medicine.”
The research team notes that most genetic variants linked to certain diseases are located in parts of the DNA that do not directly code for proteins. This makes it difficult to determine what role these variants play in someone’s health.
Lifelong Neural Network for Gene Regulation
To better understand these genetic variants and their GRNs, the researchers developed a deep learning tool called Lifelong Neural Network for Gene Regulation (LINGER). “With this tool, we aim to answer important questions, such as how and why genetic variants influence certain traits through complex GRN interactions,” says Duren. “If we can unravel these mechanisms, we may be able to predict how that person will respond to a drug based on someone’s genes. This could ultimately lead to the development of more targeted treatments.”
The researchers say that models already exist that predict how GRNs work. However, the possibilities of these models are limited because there is too little data available. “Because the data relates to one cell, the number of observations per cell is very limited,” says Duren. “But the gene regulatory network is so complex that enormous amounts of data are needed to understand it. And there just isn’t enough independent data available. We do have data from all kinds of cells, but that data is not independent.”
LINGER therefore combines external bulk data covering different cellular contexts with existing knowledge about transcription factor motifs, allowing researchers to estimate transcription factor activity and gain insight into the factors that cause disease.
Predict more accurately
“Numerous methods have been developed over the past twenty years for deriving gene regulatory networks,” says Duren. “However, our systematic benchmarks based on experimental data show that these existing models have an accuracy that is only about 17% to 29% higher than the random predictor. The new method is 125% more accurate than the random predictor.”
The research team states that LINGER could be used in several ways, not only for research into precision medicine, but also in molecular and developmental biology. In the future, the researchers want to use the tool to find more effective treatments for drug addiction.
Computer models and medication
In addition to computer models to gain insight into processes, more and more intelligent tools are also being developed to optimize or innovate medication. For example, neurologists at Erasmus MC have developed a computer model that helps them prescribe the best and cheapest MS medications to patients.
And a computer model has been developed in Utrecht with which the medication for rheumatism can be reduced and that can predict and prevent the risk of a flare-up of the complaints. This model uses data such as previous blood tests, medication use and disease activity to help doctors choose the most appropriate treatment.
