從2019年12月起新冠肺炎爆發，防疫措施愈來愈重要，隨著新型冠狀病毒的高傳播力和突變力，新的病毒株陸續出現，而這將導致疫苗防護力下降，甚至可能被突破，本研究將重點放在預測病毒株下一階段突變預測，把收集到的資料依照Nextstrain clade分類，將2021年1月以前非delta病毒出現前的DNA病毒序列提取出來，透過Variational AutoEncoder將DNA病毒序列提取其生物資訊後，使用高斯過程進行潛空間下比較不同kernel進行突變趨勢擬合，並生成之後1到6個月後的新DNA病毒序列，最後成功生成出新的病毒序列，並且突變出Dalta病毒株，說明DNA序列在Variational AutoEncoder上，成功提取其生物資訊，並能依時間演化建模，而這有助於提前開發新疫苗和預測新症狀等。;Since the outbreak of new coronary pneumonia in December 2019, epidemic prevention measures have become more and more important. With the high transmissibility and mutation power of the new coronavirus, new virus strains have emerged one after another, which will lead to a decline in vaccine protection and may even be broken through. This study will focus on predicting the next-stage mutation prediction of virus strains, and classify the collected data according to the Nextstrain clade. Using the Gaussian process to compare different kernels in the latent space to fit the mutation trend, and generate a new DNA virus sequence 1 to 6 months later, and finally successfully generate a new virus sequence, and mutate the Dalta virus strain. It shows that the biological information of the DNA sequence can be successfully extracted on the Variational AutoEncoder, and can be modeled according to the evolution of time, which is helpful for the development of new vaccines and the prediction of new symptoms in advance.
