Quantifying information of intracellular signaling: progress with machine learning

Abstract

Cells convey information about their extracellular environment to their core functional machineries. Studying the capacity of intracellular signaling pathways to transmit information addresses fundamental questions about living systems. Here, we review how information-theoretic approaches have been used to quantify information transmission by signaling pathways that are functionally pleiotropic and subject to molecular stochasticity. We describe how recent advances in machine learning have been leveraged to address the challenges of complex temporal trajectory datasets and how these have contributed to our understanding of how cells employ temporal coding to appropriately adapt to environmental perturbations.

Neural-network solutions to stochastic reaction networks

“…We have achieved such an extension, leading to the first approach of using the neural network alone to solve the chemical master equation. In light of the strong representation power of the neural network [36,37,58,59], the approach is generally applicable to stochastic reaction networks. We have applied it to representative examples in biophysics, systems biology and epidemics, where it shows a high accuracy to track the joint probability distribution over time.…”

Neural-network solutions to stochastic reaction networks

“…We have achieved such an extension, leading to the first approach of using the neural network alone to solve chemical master equation. In the light of the strong representation power of the neural network [36,37,58,59], the approach is generally applicable to stochastic reaction networks. We have applied it to representative examples in biophysics, system biology and epidemic, where it shows a high accuracy to track the joint probability distribution over time.…”

Proofreading Is Too Noisy For Effective Ligand Discrimination

“…S1 C). However, the enhancement of specificity compared to a non-proofread receptor is very moderate (less than 10-fold for Another approach to quantify the effects of noise on ligand discrimination is through the mutual information, I(n, k off ), between the input variable (ligand affinity), k off , and the receptor output, n (18,22,59). Although not the main thrust of this paper, it has been shown that mutual information is closely related to the Fisher Information which we considered in this section (66,67).…”

“…Extending the pioneering work by Berg and Purcell, (32, 34–36, 47–53) adopted the Fisher Information and the associated framework of statistical estimation theory to quantify how receptor occupancy fluctuations fundamentally limit the precision of estimating ligand concentration. Other works have investigated the specificity of response to a particular ligand using information theoretic tools (18, 22, 36, 48, 54–59). Selecting between these various approaches to account for noise in ligand processing performance depends on the question at hand.…”