-
Energy--Information Trade-off Induces Continuous and Discontinuous Phase Transitions in Lateral Predictive Coding
Authors:
Zhen-Ye Huang,
Ruyi Zhou,
Miao Huang,
Hai-Jun Zhou
Abstract:
Lateral predictive coding is a recurrent neural network which creates energy-efficient internal representations by exploiting statistical regularity in sensory inputs. Here we investigate the trade-off between information robustness and energy in a linear model of lateral predictive coding analytically and by numerical minimization of a free energy. We observe several phase transitions in the syna…
▽ More
Lateral predictive coding is a recurrent neural network which creates energy-efficient internal representations by exploiting statistical regularity in sensory inputs. Here we investigate the trade-off between information robustness and energy in a linear model of lateral predictive coding analytically and by numerical minimization of a free energy. We observe several phase transitions in the synaptic weight matrix, especially a continuous transition which breaks reciprocity and permutation symmetry and builds cyclic dominance and a discontinuous transition with the associated sudden emergence of tight balance between excitatory and inhibitory interactions. The optimal network follows an ideal-gas law in an extended temperature range and saturates the efficiency upper-bound of energy utilization. These results bring theoretical insights on the emergence and evolution of complex internal models in predictive processing systems.
△ Less
Submitted 10 January, 2024; v1 submitted 22 February, 2023;
originally announced February 2023.
-
Unveiling interpretable development-specific gene signatures in the developing human prefrontal cortex with ICGS
Authors:
Meng Huang,
Xiucai Ye,
Tetsuya Sakurai
Abstract:
In this paper, to unveil interpretable development-specific gene signatures in human PFC, we propose a novel gene selection method, named Interpretable Causality Gene Selection (ICGS), which adopts a Bayesian Network (BN) to represent causality between multiple gene variables and a development variable. The proposed ICGS method combines the positive instances-based contrastive learning with a Vari…
▽ More
In this paper, to unveil interpretable development-specific gene signatures in human PFC, we propose a novel gene selection method, named Interpretable Causality Gene Selection (ICGS), which adopts a Bayesian Network (BN) to represent causality between multiple gene variables and a development variable. The proposed ICGS method combines the positive instances-based contrastive learning with a Variational AutoEncoder (VAE) to obtain this optimal BN structure and use a Markov Blanket (MB) to identify gene signatures causally related to the development variable. Moreover, the differential expression genes (DEGs) are used to filter redundant genes before gene selection. In order to identify gene signatures, we apply the proposed ICGS to the human PFC single-cell transcriptomics data. The experimental results demonstrate that the proposed method can effectively identify interpretable development-specific gene signatures in human PFC. Gene ontology enrichment analysis and ASD-related gene analysis show that these identified gene signatures reveal the key biological processes and pathways in human PFC and have more potential for neurodevelopment disorder cure. These gene signatures are expected to bring important implications for understanding PFC development heterogeneity and function in humans.
△ Less
Submitted 15 November, 2022;
originally announced November 2022.
-
Inferring cell-specific lncRNA regulation with single-cell RNA-sequencing data in the developing human neocortex
Authors:
Meng Huang,
Jiangtao Ma,
Changzhou Long,
Junpeng Zhang,
Xiucai Ye,
Tetsuya Sakurai
Abstract:
Long non-coding RNAs (lncRNAs) are important regulators to modulate gene expression and cell proliferation in the developing human brain. Previous methods mainly use bulk lncRNA and mRNA expression data to study lncRNA regulation. However, to analyze lncRNA regulation regarding individual cells, we focus on single-cell RNA-sequencing (scRNA-seq) data instead of bulk data. Recent advance in scRNA-s…
▽ More
Long non-coding RNAs (lncRNAs) are important regulators to modulate gene expression and cell proliferation in the developing human brain. Previous methods mainly use bulk lncRNA and mRNA expression data to study lncRNA regulation. However, to analyze lncRNA regulation regarding individual cells, we focus on single-cell RNA-sequencing (scRNA-seq) data instead of bulk data. Recent advance in scRNA-seq has provided a way to investigate lncRNA regulation at single-cell level. We will propose a novel computational method, CSlncR (cell-specific lncRNA regulation), which combines putative lncRNA-mRNA binding information with scRNA-seq data including lncRNAs and mRNAs to identify cell-specific lncRNA-mRNA regulation networks at individual cells. To understand lncRNA regulation at different development stages, we apply CSlncR to the scRNA-seq data of human neocortex. Network analysis shows that the lncRNA regulation is unique in each cell from the different human neocortex development stages. The comparison results indicate that CSlncR is also an effective tool for predicting cell-specific lncRNA targets and clustering single cells, which helps understand cell-cell communication.
△ Less
Submitted 29 November, 2022; v1 submitted 15 November, 2022;
originally announced November 2022.
-
Cancer Subtyping via Embedded Unsupervised Learning on Transcriptomics Data
Authors:
Ziwei Yang,
Lingwei Zhu,
Zheng Chen,
Ming Huang,
Naoaki Ono,
MD Altaf-Ul-Amin,
Shigehiko Kanaya
Abstract:
Cancer is one of the deadliest diseases worldwide. Accurate diagnosis and classification of cancer subtypes are indispensable for effective clinical treatment. Promising results on automatic cancer subtyping systems have been published recently with the emergence of various deep learning methods. However, such automatic systems often overfit the data due to the high dimensionality and scarcity. In…
▽ More
Cancer is one of the deadliest diseases worldwide. Accurate diagnosis and classification of cancer subtypes are indispensable for effective clinical treatment. Promising results on automatic cancer subtyping systems have been published recently with the emergence of various deep learning methods. However, such automatic systems often overfit the data due to the high dimensionality and scarcity. In this paper, we propose to investigate automatic subtyping from an unsupervised learning perspective by directly constructing the underlying data distribution itself, hence sufficient data can be generated to alleviate the issue of overfitting. Specifically, we bypass the strong Gaussianity assumption that typically exists but fails in the unsupervised learning subtyping literature due to small-sized samples by vector quantization. Our proposed method better captures the latent space features and models the cancer subtype manifestation on a molecular basis, as demonstrated by the extensive experimental results.
△ Less
Submitted 2 April, 2022;
originally announced April 2022.
-
Reinforced Molecular Optimization with Neighborhood-Controlled Grammars
Authors:
Chencheng Xu,
Qiao Liu,
Minlie Huang,
Tao Jiang
Abstract:
A major challenge in the pharmaceutical industry is to design novel molecules with specific desired properties, especially when the property evaluation is costly. Here, we propose MNCE-RL, a graph convolutional policy network for molecular optimization with molecular neighborhood-controlled embedding grammars through reinforcement learning. We extend the original neighborhood-controlled embedding…
▽ More
A major challenge in the pharmaceutical industry is to design novel molecules with specific desired properties, especially when the property evaluation is costly. Here, we propose MNCE-RL, a graph convolutional policy network for molecular optimization with molecular neighborhood-controlled embedding grammars through reinforcement learning. We extend the original neighborhood-controlled embedding grammars to make them applicable to molecular graph generation and design an efficient algorithm to infer grammatical production rules from given molecules. The use of grammars guarantees the validity of the generated molecular structures. By transforming molecular graphs to parse trees with the inferred grammars, the molecular structure generation task is modeled as a Markov decision process where a policy gradient strategy is utilized. In a series of experiments, we demonstrate that our approach achieves state-of-the-art performance in a diverse range of molecular optimization tasks and exhibits significant superiority in optimizing molecular properties with a limited number of property evaluations.
△ Less
Submitted 14 November, 2020;
originally announced November 2020.
-
Reply to "Issues arising from benchmarking single-cell RNA sequencing imputation methods"
Authors:
Mo Huang,
Nancy R. Zhang
Abstract:
In our Brief Communication (DOI: 10.1038/s41592-018-0033-z), we presented the method SAVER for recovering true gene expression levels in noisy single cell RNA sequencing data. We evaluated the performance of SAVER, along with comparable methods MAGIC and scImpute, in an RNA FISH validation experiment and a data downsampling experiment. In a Comment [arXiv:1908.07084v1], Li & Li were concerned with…
▽ More
In our Brief Communication (DOI: 10.1038/s41592-018-0033-z), we presented the method SAVER for recovering true gene expression levels in noisy single cell RNA sequencing data. We evaluated the performance of SAVER, along with comparable methods MAGIC and scImpute, in an RNA FISH validation experiment and a data downsampling experiment. In a Comment [arXiv:1908.07084v1], Li & Li were concerned with the use of the downsampled datasets, specifically focusing on clustering results obtained from the Zeisel et al. data. Here, we will address these comments and, furthermore, amend the data downsampling experiment to demonstrate that the findings from the data downsampling experiment in our Brief Communication are valid.
△ Less
Submitted 5 September, 2019;
originally announced September 2019.
-
Fluctuations in Gene Regulatory Networks as Gaussian Colored Noise
Authors:
Ming-Chang Huang,
Jinn-Wen Wu,
Yu-Pin Luo,
Karen G. Petrosyan
Abstract:
The study of fluctuations in gene regulatory networks is extended to the case of Gaussian colored noise. Firstly, the solution of the corresponding Langevin equation with colored noise is expressed in terms of an Ito integral. Then, two important lemmas concerning the variance of an Ito integral and the covariance of two Ito integrals are shown. Based on the lemmas, we give the general formulae…
▽ More
The study of fluctuations in gene regulatory networks is extended to the case of Gaussian colored noise. Firstly, the solution of the corresponding Langevin equation with colored noise is expressed in terms of an Ito integral. Then, two important lemmas concerning the variance of an Ito integral and the covariance of two Ito integrals are shown. Based on the lemmas, we give the general formulae for the variances and covariance of molecular concentrations for a regulatory network near a stable equilibrium explicitly. Two examples, the gene auto-regulatory network and the toggle switch, are presented in details. In general, it is found that the finite correlation time of noise reduces the fluctuations and enhances the correlation between the fluctuations of the molecular components.
△ Less
Submitted 22 December, 2009; v1 submitted 23 October, 2009;
originally announced October 2009.
-
Potential for regulatory genetic networks of gene expression near a stable point
Authors:
Ming-Chang Huang,
Yu-tin Huang,
Jinn-Wen Wu,
Tien-Shen Chung
Abstract:
A description for regulatory genetic network based on generalized potential energy is constructed. The potential energy is derived from the steady state solution of linearized Fokker-Plank equation, and the result is shown to be equivalent to the system of coupled oscillators. The correspondence between the quantities from the mechanical picture and the steady-state fluctuations is established.…
▽ More
A description for regulatory genetic network based on generalized potential energy is constructed. The potential energy is derived from the steady state solution of linearized Fokker-Plank equation, and the result is shown to be equivalent to the system of coupled oscillators. The correspondence between the quantities from the mechanical picture and the steady-state fluctuations is established. Explicit calculation is given for auto-regulatory networks in which, the force constant associated with the degree of protein is very weak. Negative feedback not only suppresses the fluctuations but also increases the steepness of the potential. The results for the fluctuations agree completely with those obtained from linear noise Fokker-Planck equation.
△ Less
Submitted 20 October, 2007;
originally announced October 2007.
