Finding gene network topologies for given biological function with recurrent neural network

Abstract

Searching for possible biochemical networks that perform a certain function is a challenge in systems biology. For simple functions and small networks, this can be achieved through an exhaustive search of the network topology space. However, it is difficult to scale this approach up to larger networks and more complex functions. Here we tackle this problem by training a recurrent neural network (RNN) to perform the desired function. By developing a systematic perturbative method to interrogate the successfully trained RNNs, we are able to distill the underlying regulatory network among the biological elements (genes, proteins, etc.). Furthermore, we show several cases where the regulation networks found by RNN can achieve the desired biological function when its edges are expressed by more realistic response functions, such as the Hill-function. This method can be used to link topology and function by helping uncover the regulation logic and network topology for complex tasks.

Inferring structural and dynamical properties of gene networks from data with deep learning

“…However, for more complex system, the learned f i is not necessarily monotonous, but has a parabolic shape ( Supplementary Figures S4B and S5B ). So we introduce an edge removal strategy to infer the network structure ( 25 ). We assume that there exist interactions between any node pair, and the strength of the interaction is represented by the network connection weight.…”

“…Recently, Shen et al. have used recurrent neural network (RNN) to infer the structure of gene regulatory networks successfully, in several biological systems with different functions including adaptation, controlled oscillation, and pattern formation ( 25 ). This work emphasizes the potential effectiveness of deep neural network (DNN) used for gene network reconstruction.…”

“…On the other hand, the reconstruction of gene networks from single cell data is another challenging problem ( 26–28 ). Inspired by previous studies on DNN and the development of neural ordinary differential equations (neural ODEs) ( 25 , 29–32 ), we seek to apply DNN to multistable and oscillation systems for the network reconstruction. Especially, we consider the ensemble of multiple models to extract more information from the data to address the unique challenges posed by single-cell data, such as cell-to-cell variabilities, short time series length, and high sparsity.…”

Engineering synthetic spatial patterns in microbial populations and communities

“…To engineer such multistep patterning programs, the field will benefit from recent and future advances in synthetic biology, such as the use of artificial intelligence and whole-cell simulations [75]. In the future, machine learning may be implemented to design intricate gene-regulatory networks for a desired output [76].…”

How artificial intelligence enables modeling and simulation of biological networks to accelerate drug discovery

“…Open issues in model construction include the quality of the reconstructed biological networks (eventually including epigenomic data), because many genes have an unknown function and the literature is often ambiguous or incomplete. In this regard, AI can be used to support/validate the prediction of gene network topology starting from a desired biological function (Shen et al, 2021). Another outstanding challenge is the development of automatic and reliable computational methods for the prediction of the baseline values of concentrations and fluxes, something that as discussed necessarily goes along with the improvement of experimental high-throughput methods and availability of transparent highquality multi-omics data.…”