ProAll-D

Allergy, often described as an autoimmune disorder, is a clinical condition characterized by the sensitivity of the immune system to normally innocuous elements. Breathing issues, food sensitivities, eczema, bronchopneumonia, and hypersensitivity reaction are examples of these conditions. Symptoms also include red eyes, itching, coughing, nasal congestion, breathing difficulty, or swelling. Typical allergens include pollen and certain foods. Food, insect bites, and drugs are all major causes of severe reactions. Their growth is influenced by both hereditary and environmental factors. Bioinformatics has been extensively utilized to predict allergens, involving a range of tools and methodologies. Even though discriminative algorithms such as support vector machines (SVM) have recently improved alignment methods, they are time-consuming if a new pattern must be compared to a huge amount of existing data. Another limitation of the alignment methods is that they won't develop a model of the true positive but rather estimate the average distances between protein sequences. We present a set of novel allergen prediction models based on protein characteristic values, ACC (auto and cross-covariance) transformation, and a variety of supervised learning, ensemble learning, and deep learning classification methods such as Gaussian Nave Bayes, Radius Neighbour’s Classifier, Extra Tree Classifier, Bagging Classifier, ADA Boost, Linear Discriminant Analysis, Quadratic Discriminant Analysis, and LSTM methods. LSTM was the best-performing model, with an accuracy of 91.51 percent.过敏（Allergy）常被视为一种自身免疫性疾病，是指免疫系统对原本无害的物质产生异常敏感反应的临床病症。此类病症的表现包括呼吸障碍、食物不耐受、湿疹、支气管肺炎以及超敏反应等，相关症状还包括眼结膜发红、瘙痒、咳嗽、鼻塞、呼吸困难或组织肿胀等。典型过敏原（allergen）包括花粉与特定食物，而食物、昆虫叮咬以及药物则是引发严重超敏反应的主要诱因。过敏的发生发展同时受遗传与环境因素的共同影响。生物信息学（Bioinformatics）已被广泛应用于过敏原预测研究，涵盖了诸多工具与研究方法。尽管诸如支持向量机（SVM）这类判别式算法近期优化了序列比对方法，但当需要将新的序列模式与海量现有数据进行比对时，其计算耗时较长。序列比对方法的另一局限在于，它们无法构建真阳性样本的预测模型，仅能估算蛋白质序列之间的平均距离。本研究提出了一套基于蛋白质特征值、自协方差与互协方差（ACC，auto and cross-covariance）变换，以及多种监督学习、集成学习与深度学习分类方法的新型过敏原预测模型，所涉方法包括高斯朴素贝叶斯、半径邻居分类器、极端随机树分类器、装袋分类器、自适应提升算法、线性判别分析、二次判别分析以及长短期记忆网络（LSTM）等。其中长短期记忆网络（LSTM）的模型性能最优，分类准确率可达91.51%。