【摘要】 提出了一个由7个BP神经网络组合成的多模神经网络的预测模型,同时给多模神经网络引进了较多的生物进化信息(Evolutionary information),即一方面引入了"profile"编码,这种编码被认为携带了较多的生物信息;另一方面引入了氨基酸之间的"距离"概念.它体现了输入层临近氨基酸的相互联系和影响.对从36个蛋白质提取的4 000个氨基酸的进行了预测研究.结果表明,与文献[1]的预测结果相比,本文的多模神经网络把蛋白质二级结构预测的平均精度从66.1502％提高到68.8903％.更多还原

【Abstract】 A multi-modal neural network that was made of seven feed - forward BP neural networks to predict the secondary structure of proteins is developed. And the more biological evolution information into this multi modal neural network,that is to say"profile" code is introduced,which is thought to carry more evolution information, and on the other hand, the "distance" concept between the amino acid is introduced.It has embodied the connection and influence of importing layer close to amino acid. A prediction is made on the protein secondary structure by using 4000 amino acids from 36 proteins. Results indicate that as compared with[1]whose result is 66.1502% ,our Multi-modal Neural Network may increase the average accuracy to 68.890 3%.更多还原