摘要： 为使得隐马尔可夫模型(HMM)能够处理非相邻可见符号之间的依赖关系，将延时机制引入标准的HMM中。该技术仅仅改变了高阶状态发射概率的计算。所有适用于HMM的算法基本保持不变。该文设计了一个一阶延时隐马尔可夫模型和一个一阶标准隐马尔可夫模型，将两者分别应用于水稻基因剪接供体位点的识别。识别结果显示，延时模型的判别能力在一定程度上优于标准模型。对那些特征很不符合的位点，延时模型给出了相对低得多的得分。

关键词: 隐马尔可夫模型, 延时, 剪接供体位点识别, 水稻基因组

Abstract: To enable hidden Markov models to account for dependencies between non-adjacent observation symbols, time-delay is introduced to standard high order HMM states. This technique only changes the calculation of emission probabilities in high order states. All the algorithms for HMM remains almost the same. Such a time-delay first order HMM as well as a standard first order HMM is established for splice donor sites in rice genome. The results show some improvements in discriminative power for time-delay first order HMM vs standard first order HMM. It is worth noting that the former gives much lower scores to sites with poor potential as donor signals from the remainder of sites.

Key words: Hidden Markov model (HMM), Time-delay, Splice donor sites prediction, Rice genome