Synthetic Generation of Dynamic Omics Data Demonstrates Aspergillus nidulans BrlA Paradoxical Wall Stress Response

We propose a method to generate additional dynamic omics trajectories which could support pathway analysis methods such as enrichment analysis, genetic programming, and machine learning. Using long short-term memory neural networks, we can effectively predict an organisms dynamic response to a stimulus based on an initial dataset with relatively few samples. We present both an in silico proof of principle, based on a model that simulates viral propagation, and an in vitro case study, tracking the dynamics of Aspergillus nidulans BrlA transcript in response to antifungal agent micafungin. Our silico experiment was conducted using a highly noisy dataset with only 25 replicates. This proof of principle shows that this method can operate on biological datasets, which often have high variance and few replicates. Our in silico validation achieved a maximum R2 value of approximately 0.95 on our highly noisy, stochastically simulated data. Our in vitro validation achieves an R2 of 0.71. As with any machine learning application, this method will work better with more data; however, both of our applications attain acceptable validation metrics with very few biological replicates. The in vitro experiments also revealed a novel paradoxical dose-response effect: transcriptional upregulation by Aspergillus nidulans BrlA is highest at an intermediate dose of 10 ng/mL and is reduced at both higher and lower concentrations of micafungin.