Machine Learning for Classification and Genome Annotations

这个名为“分类和基因组注释的机器学习”的提案建立在我之前由NSERC资助的名为“计算智能和生物信息学应用”的研究项目的基础上。计算生物学中的许多问题涉及分类，例如将细胞样本分为恶性（癌症）或良性（正常）。对于这样的任务，可以开发和使用神经网络。神经网络为机器学习提供了一个很好的通用框架，包括分类、图像分割和理解。虽然神经网络领域目前是一个热门的研究领域，在图像处理和计算机视觉方面取得了许多成功，但对单个层或节点的激活函数学习的关注相对较少。为此，我们提出了EvoDNN，这是一种进化深度神经网络，它采用进化算法来进化深度异构前馈神经网络。当前大多数前馈神经网络采用用户定义的同构激活函数，而EvoDNN创建了异构多层网络，其中每个神经元的激活函数不是由用户静态定义的，而是在进化过程中动态优化的。EvoDNN提供的一个优点是激活函数不需要是可微的。这个特性为用户提供了很大程度的灵活性，可以选择使用哪些激活函数。2)半自动基因组注释的机器学习方法（SAGA）本提案的第二个主题与基因组注释领域的机器学习有关。半自动基因组注释（Semi - automated genome annotation， SAGA）方法被广泛用于了解基因组活性和基因调控。这些方法将一组基于测序的表观基因组活性测定（如组蛋白修饰和转录因子结合的ChIP-seq测量）作为输入，并输出基因组的注释，该注释将染色质状态标签分配给每个基因组位置。由于离散注释框架，现有的SAGA方法存在一些局限性：这种注释不能很容易地表示基因组元素的不同强度，也不能很容易地表示同时表现出多种活动类型的组合元素。为了弥补这些限制，我们提出了一种注释策略，在每个位置输出染色质状态特征的向量，而不是单个离散标签。连续建模在其他领域也很常见，比如文本文档的主题建模。我们提出了一种方法，即表观基因组-ssm，该方法使用非负卡尔曼滤波状态空间模型有效地用染色质状态特征注释基因组。我们打算表明，与连续和离散替代方案相比，表观基因组-ssm的染色质状态特征对几种下游应用更有用。

This proposal entitled "Machine Learning for Classification and Genome Annotations" builds on my previous research program funded by NSERC entitled "Computational Intelligence and Applications in Bioinformatics". It has 2 components: 1) Self-adjusting Evolutionary Neural Networks and Applications in Computational Biology Many problems in Computational Biology involve classification, such as classifying cell samples into malignant (cancer) or benign (normal). For such tasks, Neural Networks can be developed and employed. Neural Networks offer a great general framework for machine learning, including classification, image segmentation and understanding. While the neural network field is currently a hot research area, with many successes in image processing and computer vision, relatively little attention has been paid to the learning of activation functions for individual layers or nodes. For this purpose, we propose EvoDNN, an evolutionary deep neural network that employs an evolutionary algorithm to evolve deep heterogeneous feed-forward neural networks. While the majority of current feed-forward neural networks employ user defined homogeneous activation functions, EvoDNN creates heterogeneous multi-layer networks where each neuron's activation function is not statically defined by the user, but dynamically optimized during evolution. One advantage offered by EvoDNN is that the activation functions do not need to be differentiable. This feature gives users a great degree of flexibility over which activation functions to employ. 2) Machine Learning Methods for Semi Automated Genome Annotation (SAGA) A second topic of this proposal is related to machine learning in the area of genome annotation. Semi automated genome annotation (SAGA) methods are widely used to understand genome activity and gene regulation. These methods take as input a set of sequencing based assays of epigenomic activity (such as ChIP-seq measurements of histone modification and transcription factor binding), and output an annotation of the genome that assigns a chromatin state label to each genomic position. Existing SAGA methods have several limitations caused by the discrete annotation framework: such annotations cannot easily represent varying strengths of genomic elements, and they cannot easily represent combinatorial elements that simultaneously exhibit multiple types of activity. To remedy these limitations, we propose an annotation strategy that instead outputs a vector of chromatin state features at each position rather than a single discrete label. Continuous modeling is common in other fields, such as in topic modeling of text documents. We propose a method, epigenome-ssm, that uses a nonnegative Kalman filter state space model to efficiently annotate the genome with chromatin state features. We intend to show that chromatin state features from epigenome-ssm are more useful for several downstream applications than both continuous and discrete alternatives.