Abstract
To survive environmental conditions, cells transcribe their response activities into encoded mRNA sequences in order to produce certain amounts of protein concentrations. The external conditions are mapped into the cell through the activation of special proteins called transcription factors (TFs). Due to the difficult task to measure experimentally TF behaviors, and the challenges to capture their quick-time dynamics, different types of models based on differential equations have been proposed. However, those approaches usually incur in costly procedures, and they present problems to describe sudden changes in TF regulators. In this paper, we present a switched dynamical latent force model for reverse-engineering transcriptional regulation in gene expression data which allows the exact inference over latent TF activities driving some observed gene expressions through a linear differential equation. To deal with discontinuities in the dynamics, we introduce an approach that switches between different TF activities and different dynamical systems. This creates a versatile representation of transcription networks that can capture discrete changes and non-linearities. We evaluate our model on both simulated data and real data (e.g., microaerobic shift in E. coli, yeast respiration), concluding that our framework allows for the fitting of the expression data while being able to infer continuous-time TF profiles.
Introduction
In systems biology, "reverse-engineering" refers to the task of identifying the hidden control signals—Transcription Factors (TFs)—that drive observed outputs like mRNA levels. While mRNA levels can be measured relatively easily using microarrays or RNA-seq, measuring the real-time activity of TFs is often technically impossible or prohibitively expensive.
Existing models for this task frequently suffer from two major drawbacks. First, they often assume that TF activity changes smoothly over time, failing to capture the "switches"—sudden metabolic shifts triggered by environmental stressors like oxygen depletion. Second, models capable of handling such discontinuities often rely on slow, computationally intensive sampling methods like Markov Chain Monte Carlo (MCMC). This paper introduces the Switched Latent Force Model (SLFM), a framework that is mathematically rigorous enough to handle discontinuities while remaining efficient enough for exact inference.
The Switched Latent Force Model (SLFM)
The SLFM combines Ordinary Differential Equations (ODEs) with Gaussian Processes (GPs) to create a "physics-aware" statistical model. The core process is described by a first-order ODE representing the production (transcription) and degradation (decay) of mRNA. Instead of using a generic curve-fitting tool, the authors derive a custom covariance function (kernel) directly from this ODE.
The primary innovation is the "switch" mechanism. The timeline is divided into non-overlapping segments. Within each segment, the system can be driven by different TF profiles or operate under different biological parameters. A continuity constraint is enforced at the boundaries: while the latent TF activity ($u(t)$) can jump instantaneously, the resulting mRNA levels ($z(t)$) must remain continuous, reflecting physical reality.
By embedding the ODE into the GP kernel, the model "understands" the physical constraints of transcription and decay. This provides a natural regularization that generic machine learning models lack, allowing the SLFM to identify the exact moment a cell changes its metabolic strategy without "blurring" sharp transitions.
Experimental Validation and Results
The model was validated across three distinct datasets: synthetic "toy" data, E. coli microaerobic shift data, and yeast respiration cycles.
In the E. coli study, the model tracked five genes during a transition from aerobic to anaerobic conditions. It successfully identified the activation of the FNR regulator approximately 15–20 minutes after oxygen depletion, outperforming smooth models that struggled with the suddenness of the shift. In the yeast metabolic cycle data, the SLFM captured the periodic nature of the cycle across 36 time points. It revealed that the RAP1 protein has a significantly stronger influence (roughly 5x) on ribosomal production than FHL1, a finding consistent with biological literature.
The SLFM demonstrated superior accuracy in identifying the timing of TF activation compared to previous state-of-the-art switching models. Its ability to handle "bang-bang" control—where biological systems operate like a thermostat, switching between on and off states—makes it a powerful tool for capturing discrete changes and non-linearities in gene regulation.
Conclusion
The Switched Latent Force Model provides a versatile and efficient framework for reverse-engineering transcriptional networks. By allowing for exact GP inference, it avoids the computational bottlenecks of stochastic simulations while maintaining the ability to capture sudden biological shifts. Future work will focus on automating the selection of time segments and latent forces, as well as extending the framework to handle non-linear biological interactions and genome-wide datasets through sparse GP approximations.
Summary updated