Mathematically Rigorous Results In Sequence Matching

序列匹配中严格的数学结果

• 批准号: 7148038
• 负责人: John L Spouge
• 金额: --
• 依托单位: NATIONAL LIBRARY OF MEDICINE
• 依托单位国家: 美国
• 资助国家: 美国
• 起止时间: 至
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/7148038
• 关键词: artificial intelligence; computer assisted sequence analysis; computer simulation; mathematics; nucleic acids; statistics /biometry

The BLAST programs (BLASTP, PSI-BLAST, etc.) presently use offline computer simulations to give accurate estimates of statistical significance for sequence matches. This project has already speeded those offline computer simulations by a factor of 100-1000. Its eventual aim is to speed the simulations further, so they can be done online over the web. If the project is successful, BLAST users will then be free to use any scores and penalties they choose for matching sequences. There are two parameters in sequence matching statistics: the scale parameter ?lambda? and the pre-factor ?k?. We heuristically derived the new equation for scale parameter ?lambda?. This equation can estimate ?lambda? efficiently with high accuracy. In addition, we have proposed the new formula for Gumbel pre-factor ?k? based on the path reversal identity and the Poisson clumping heuristic. This formula also provides very accurate results. In our next step, we will explore edge effects on the statistics. Edge effects are present because real sequences have limited lengths appear as a correction term in an asymptotic expansion of the probability of sequence matching. This edge effect is likely to be more important in the statistics of matching with gaps than it was in the statistics of matching without gaps, because gapped matches tend to be longer, exhausting the sequences being matched more easily.

BLAST 程序（BLASTP、PSI-BLAST 等）目前使用离线计算机模拟来准确估计序列匹配的统计显着性。该项目已经将离线计算机模拟的速度提高了 100-1000 倍。其最终目标是进一步加快模拟速度，以便可以通过网络在线完成模拟。如果该项目成功，BLAST 用户将可以自由使用他们为匹配序列选择的任何分数和惩罚。序列匹配统计中有两个参数：尺度参数？lambda？和前置因子 ?k? 。我们启发式地导出了尺度参数“lambda”的新方程。这个方程可以估计 lambda？高效、高精度。此外，我们还提出了 Gumbel 前置因子 ?k? 的新公式。基于路径反转恒等式和泊松聚类启发式。该公式还提供了非常准确的结果。在下一步中，我们将探讨统计数据的边缘效应。存在边缘效应是因为真实序列的长度有限，作为序列匹配概率的渐近展开中的校正项出现。这种边缘效应在有间隙匹配的统计中可能比在无间隙匹配的统计中更重要，因为有间隙的匹配往往更长，更容易耗尽正在匹配的序列。