Abstract:Gene expression is inherently dynamic, due to complex regulation and stochastic biochemical events. However, the effects of these dynamics on cell phenotypes can be difficult to determine. Researchers have historically been limited to passive observations of natural dynamics, which can preclude studies of elusive and noisy cellular events where large amounts of data are required to reveal statistically significant effects. Here, using recent advances in the fields of machine learning and control theory, we train a deep neural network to accurately predict the response of an optogenetic system in Escherichia coli cells. We then use the network in a deep model predictive control framework to impose arbitrary and cell-specific gene expression dynamics on thousands of single cells in real time, applying the framework to generate complex time-varying patterns. We also showcase the framework's ability to link expression patterns to dynamic functional outcomes by controlling expression of the tetA antibiotic resistance gene. This study highlights how deep learning-enabled feedback control can be used to tailor distributions of gene expression dynamics with high accuracy and throughput without expert knowledge of the biological system.

What problem does this paper attempt to address?

The core problem that this paper attempts to solve is: how to achieve precise dynamic regulation of single - cell gene expression through deep learning and model predictive control techniques. Specifically, the researchers aim to overcome the limitations of traditional methods in the dynamic study of single - cell gene expression, especially the problems of insufficient data volume and the lack of effective tools to generate arbitrary gene expression dynamics. ### Specific Background of the Problem 1. **Dynamic Characteristics of Gene Expression** - Gene expression is dynamically changing and is affected by complex regulatory mechanisms and random biochemical events. - The impact of these dynamic changes on cell phenotypes is difficult to determine, especially in cases where a large amount of data is required to reveal statistically significant effects. 2. **Limitations of Existing Research** - Previous studies have been mainly limited to passively observing naturally occurring gene expression dynamics, which makes it difficult to study those elusive and noisy cell events. - Studies at single - cell resolution (such as genomics, transcriptomics, and flow cytometry) can quantify expression data with high throughput, but these measurements are usually "snapshots" and cannot capture the time context of cells. 3. **Lack of Tools for Generating Arbitrary Gene Expression Dynamics** - Existing tools are difficult to generate precise gene expression dynamics, especially when cells show different expression patterns under the same optogenetic input. ### Solutions in the Paper To solve the above problems, the researchers took advantage of the latest advances in machine learning and control theory and developed a method based on Deep Model Predictive Control (DMPC). The specific steps are as follows: 1. **Establishment of the Experimental Platform** - A high - throughput experimental platform was constructed, which can culture, observe, and optogenetically stimulate Escherichia coli cells at the single - cell level. - The CcaSR optogenetic system was used, with 535 nm green light activating gene expression and 670 nm red light inhibiting gene expression. 2. **Training of the Deep Learning Model** - A deep neural network was trained to accurately predict the response of the optogenetic system. - A deep learning model with an encoder - decoder structure, combined with Long Short - Term Memory (LSTM) networks and Multi - Layer Perceptrons (MLP), was used to predict future gene expression dynamics. 3. **Real - Time Feedback Control** - The trained deep learning model was applied to a deep model predictive control framework to achieve real - time feedback control. - Customized optogenetic stimuli were independently applied through a Digital Micromirror Device (DMD), enabling thousands of single cells to generate arbitrary and specific gene expression dynamics. 4. **Functional Verification** - This framework was applied to control the expression of the antibiotic - resistance gene tetA, generating complex time - varying patterns, and linking the expression patterns to dynamic functional results. - Through high - resolution data analysis, the relationship between gene expression levels, growth rates, and survival was revealed. ### Conclusion This study shows that deep model predictive control can be used to precisely regulate single - cell gene expression dynamics, with the characteristics of high precision and high throughput. This method can not only generate arbitrary gene expression patterns but also reveal the causal relationship between gene expression dynamics and cell functions, thus providing new tools and ideas for in - depth understanding of complex biological systems.

Deep model predictive control of gene expression in thousands of single cells

Long-term model predictive control of gene expression at the population and single-cell levels

Deep Neural Networks for Predicting Single-Cell Responses and Probability Landscapes

Functional Inference of Gene Regulation Using Single-Cell Multi-Omics

Biologically Informed Deep Learning to Infer Gene Program Activity in Single Cells

In-Vivo Real-Time Control of Protein Expression from Endogenous and Synthetic Gene Networks

A computational framework for optimal and Model Predictive Control of stochastic gene regulatory networks

Multi-Experiment Nonlinear Mixed Effect Modeling of Single-Cell Translation Kinetics after Transfection

Controlling gene expression with deep generative design of regulatory DNA

Learning the Regulatory Code of Gene Expression

Automated optogenetic feedback control for precise and robust regulation of gene expression and cell growth

Learning interpretable cellular responses to complex perturbations in high-throughput screens

Deciphering gene regulation from gene expression dynamics using deep neural network

Parameterizing Cell-to-cell Regulatory Heterogeneities Via Stochastic Transcriptional Profiles

Deep Learning and Computer Vision Strategies for Automated Gene Editing with a Single-Cell Electroporation Platform

A Machine Learning Approach to Simulate Gene Expression and Infer Gene Regulatory Networks

Mathematical modeling of RNA-based architectures for closed loop control of gene expression.

A genome-scale deep learning model to predict gene expression changes of genetic perturbations from multiplex biological networks

A deep generative model for single-cell RNA sequencing with application to detecting differentially expressed genes

Cell reprogramming design by transfer learning of functional transcriptional networks

Data-Driven Prediction of CRISPR-Based Transcription Regulation for Programmable Control of Metabolic Flux