NEW YORK—Researchers at the Icahn School of Medicine at Mount Sinai and the University of Washington have designed a modeling system that integrates genomic and temporal information to infer causal relationships between genes, drugs and their environment, allowing for a more accurate prediction of their interactions over time. The work is described in a paper entitled “Temporal Genetic Association and Temporal Genetic Causality Methods for Dissecting Complex Networks,” which was published Sept. 28 in Nature Communications.
Given the complexity of biological systems, researchers believed it would only be possible to increase accuracy of prediction tools by examining gene expression and other data in response to various perturbations at multiple points over time. The tools they created measure both static and dynamic changes, in order to identify the web of causal relationships among molecular elements that make up regulatory networks.
“In general, genes X drugs/environment interactions are studied at one most informative time point, which is predefined based on some knowledge. However, most responses to drugs or environment are dynamic. It is hard to find a universal most informative time point for all possible genes X drugs/environment interactions,” mentions Dr. Jun Zhu, a professor of genetics and genomic sciences at the Icahn School of Medicine, Head of Data Sciences at Sema4 and senior author of the publication. “Instead, modeling these interactions at one specific time point, we proposed to leverage time series data and developed a temporal genetic association testing method to model genes X drugs/environment interactions.
“Responses of gene expression traits over time are very diverse, so we used a polynomial function (of time) to describe temporal trajectories and further assumed that the temporal traits follow a multivariate normal distribution with a flexible covariance structure across subsequent time points. Then, we can test the association between the temporal traits and genetic loci with or without drugs. In addition, we developed a causality test to distinguish associations vs. causal regulations, aiming to discover molecular mechanisms of model genes X drugs/environment interactions.”
The scientists evaluated their tools by analyzing a genetically heterogenous population of yeast cells treated with rapamycin, a potential anti-aging drug, profiling the population at multiple time points.
“Yeast is a well-studied model organism and is commonly used in systems genetic studies. It grows very fast. Comparing with the vastly complex human biological system, a yeast system provides a relatively simpler starting point to develop and evaluate the novel temporal-genetic methods proposed in our study,” Zhu notes.
According to Zhu, “Aging and aging related diseases are one of the big healthcare issues that we are facing now. Rapamycin has been proposed as a potential anti-aging treatment, as it has been shown to extend the lifespan in mice. However, different tissues respond to rapamycin differently, and chronic usage of rapamycin can lead to insulin resistance in some cases. Therefore, to develop a rapamycin-based anti-aging treatment, it is important to understand the underlying molecular and physiological mechanisms of a rapamycin treatment, especially the gene-by-rapamycin interactions.”
The results demonstrated that the new approach identified a significant amount of associations between DNA variation and gene expression variation, especially for aging-related genes, reflecting the changing impact of genetic variations over time. This approach proved more reliable in identifying causal regulators of gene-drug interactions, compared to conventional methods using a single time point.
“Genetic gene expression regulations are not randomly distributed, and are normally grouped in a few genetic loci noted as eQTL hot spots. There were hotspots that were only identified by our temporal-genetic association method, not by other conventional methods,” Zhu continues. “Furthermore, we provided additional insights into the biological mechanisms underlying the gene-by-rapamycin interactions by inferring the causal regulators of these temporal QTL hotspots.”
Zhu also adds that the model has “a great utility in precision medicine for developing drug biomarkers or drug combinations. As mentioned above, our temporal-genetic association model is not only more sensitive in identifying genes X drugs interactions than traditional single time point-based methods, which rely on predefining time points, but is also more flexible. Traits or samples can be measured at a non-fixed series of time points. Comparing with single gene mutation-based methods (siRNA or CRISPR screening), our model has potential to identify multiple genes functioning together to interact with drugs. With identified genes X drugs interactions, drug combinations can be developed to enhance or minimize the genes X drugs interactions.
“Alternatively, it can be extended to study the effects of diverse perturbations on temporal traits—for example, virus-specific effects or vaccine responsiveness, as we have already implemented in another study. In addition, time scale also helps to dissect the causal relationship among different temporal traits to further illustrate the underlying molecular mechanisms and yield novel therapeutic targets.”
Asked about future plans for the algorithm, Zhu says, “We are generating human temporal data in responding to experimental perturbations so that we can refine temporal genetic association and causal testing methods based on diploid system instead of haploid system in yeast. We are also extending the proposed methods to study diverse perturbations—for example, virus-specific effects among diverse strains of viruses on temporal gene expression trajectories from mouse lung or blood. This extension enables us to identify causal molecular mechanisms underlying virulent strains or genes X virus interactions so that we can develop therapies to modulate host response to a specific virus. In the future, as more data is accumulated, more complicated models will be considered to represent the comprehensive regulation relationships.”
“The temporal-genetic models enhance the power to resolve causal relationships and provide a more systematic view of the dynamic regulation mechanisms. The models can be applied to many areas, such as early cancer or other diseases detection, and precision therapeutic development. The same framework can be used to study temporal gene-by-drug effects, which has the potential to greatly enhance novel drug development tailored for patients given their genetic information,” Zhu concludes. “There are lots of directions to push.”
