Phillips Chelsea A, Reading Benjamin J, Livingston Matthew, Livingston Kimberly, Ashwell Chris M
Prestage Department of Poultry Science, North Carolina State University, Raleigh, NC, United States.
Department of Applied Ecology, North Carolina State University, Raleigh, NC, United States.
Front Physiol. 2020 Feb 25;11:101. doi: 10.3389/fphys.2020.00101. eCollection 2020.
The muscle myopathy wooden breast (WB) has recently appeared in broiler production and has a negative impact on meat quality. WB is described as hard/firm consistency found within the pectoralis major (PM). In the present study, we use machine learning from our PM and liver transcriptome dataset to capture the complex relationships that are not typically revealed by traditional statistical methods. Gene expression data was evaluated between the PM and liver of birds with WB and those that were normal. Two separate machine learning algorithms were performed to analyze the data set including the sequential minimal optimization (SMO) of support vector machines (SVMs) and Multilayer Perceptron (MLP) Artificial Neural Network (ANN). Machine learning algorithms were compared to identify genes within a gene expression data set of approximately 16,000 genes for both liver and PM, which can be correctly classified from birds with or without WB. The performance of both machine learning algorithms SMO and MLP was determined using percent correct classification during the cross-validations. By evaluating the WB transcriptome datasets by 5× cross-validation using ANNs, the expression of nine genes ranked based on Shannon Entropy (Information Gain) from PM were able to correctly classify if the individual bird was normal or exhibited WB 100% of the time. These top nine genes were all protein coding and potential biomarkers. When PM gene expression data were evaluated between normal birds and those with WB using SVMs they were correctly classified 95% of the time using 450 of the top genes sorted ranked based on Shannon Entropy (Information Gain) as a preprocessing step. When evaluating the 450 attributes that were 95% correctly classified using SVMs through Ingenuity Pathway Analysis (IPA) there was an overlap in top genes identified through MLP. This analysis allowed the identification of critical transcriptional responses for the first time in both liver and muscle during the onset of WB. The information provided has revealed many molecules and pathways making up a complex molecular mechanism involved with the progression of wooden breast and suggests that the etiology of the myopathy is not limited to activity in the muscle alone, but is an altered systemic pathology.
肌肉肌病“木胸”（WB）最近出现在肉鸡生产中，对肉质有负面影响。WB的特征是胸大肌（PM）质地坚硬。在本研究中，我们利用PM和肝脏转录组数据集进行机器学习，以捕捉传统统计方法通常无法揭示的复杂关系。评估了患有WB的鸡和正常鸡的PM与肝脏之间的基因表达数据。执行了两种独立的机器学习算法来分析数据集，包括支持向量机（SVM）的序列最小优化（SMO）和多层感知器（MLP）人工神经网络（ANN）。比较机器学习算法，以在肝脏和PM的约16000个基因的基因表达数据集中识别能够正确区分有或无WB的鸡的基因。在交叉验证期间，使用正确分类百分比来确定SMO和MLP这两种机器学习算法的性能。通过使用人工神经网络进行5倍交叉验证来评估WB转录组数据集，基于PM的香农熵（信息增益）排名的9个基因的表达能够100%正确区分个体鸡是正常还是患有WB。这前九个基因均为蛋白质编码基因且是潜在的生物标志物。当使用支持向量机在正常鸡和患有WB的鸡之间评估PM基因表达数据时，作为预处理步骤，使用基于香农熵（信息增益）排序的前450个基因，95%的情况下能正确分类。当通过通路分析（IPA）评估使用支持向量机95%正确分类的450个属性时，通过MLP识别的前几个基因存在重叠。该分析首次确定了WB发病过程中肝脏和肌肉中的关键转录反应。所提供的信息揭示了许多构成与木胸进展相关的复杂分子机制的分子和通路，并表明该肌病的病因不仅限于肌肉中的活动，而是一种改变的全身病理学。
