基於變分自動編碼器之解糾纏模型設計與應用:單細胞RNA定序之聚類與細胞擾動之預測

Variational autoencoder based disentangle model design and application: scRNA-seq clustering and cell perturbation prediction

陳則光(Chen Ze-Guang)

指導教授 ： 葉家宏 ； 康立威

國立臺灣師範大學/科技與工程學院/電機工程學系/碩士(2022年)

https://doi.org/10.6345/NTNU202201313

摘要

在今天,深度網路已經應用於許多領域，包含產業界以及科學研究中，然而雖然深度網路可以自動生成出許多特徵已擬合出我們要求的結果，這些特徵卻難以被人類解讀。 當模型得出結果時,我們往往難以理解其是如何得出該結果以進行驗證其合理性，本研究的目標為設計可生成出更有解釋性特徵的基於變分自動編碼器模型，首先我們提出了可估計模型生成的特徵間的訊息相關性的方法，並藉此調控訓練過程中的超參數以使模型生成彼此訊息相互獨立的解糾纏特徵，並證明了使用這些解糾纏特徵可有效提升單細胞RNA定序的聚合正確度，本論文也提出了透過解開擾動不變訊息以預測細胞經擾動後的狀態，實驗證明這不只可以提升預測準確度，而且可以提供預測的根據，並可在某種程度上預測細胞經擾動前的狀態。

關鍵字

none

並列摘要

Today, deep models have been used in many fields, including industry and scientific research. However, although deep models can automatically generate many features that fit the results we require, these features are difficult for humans to interpret. When a result is obtained, it is often difficult for us to understand if the predicted result is reasonable. The goal of this research is to design VAE based model that can generate more explanatory features. First, we propose a method that can estimate the information correlation between the features generated by the model, and adjust the hyperparameters in the training process to make the model generate disentangled features that are independent of each other. And proved using those features can effectively improve the aggregation accuracy of single-cell RNA sequencing. This paper also proposes a model to predict the state of cells after perturbation by unraveling the perturbation invariant information. Experiments show that this can not only improve the prediction accuracy, but also provide a basis for prediction, and to some extent predict the state of cells before perturbation.

並列關鍵字

none

參考文獻

[1]X. Chen, Y. Duan, R. Houthooft, J. Schulman, I. Sutskever, and P. Abbeel, “Infogan: Interpretable representation learning by information maximizing generative adversarial nets,” Advances in neural information processing systems, vol. 29, 2016

[2]R. T. Q. Chen, X. Li, R. B. Grosse, and D. K. Duvenaud, “Isolating Sources of Disentanglement in Variational Autoencoders,” Advances in Neural Information Processing Systems, vol. 31, 2018

[3]Z. Xu, Z. Liu, C. Sun, K. Murphy, W. T. Freeman, J. B. Tenenbaum, and J. Wu, “Unsupervised Discovery of Parts, Structure, and Dynamics,” arXiv preprint arXiv:1903.05136, 2020

[4]H. Yu, and J. D. Welch, “MichiGAN: sampling from disentangled representations of single-cell data using generative adversarial networks,” Genome Biology, pp. 1-26, 2021

[5]D. P. Kingma, and M. Welling, “Auto-encoding variational bayes,” arXiv preprint arXiv:1312.6114.