北京雁栖湖应用数学研究院

丘成栋

胡国庆

2024年11月04日 10:00 至 10:30

Online

In this report, we introduce a novel approach based on the Asymmetric Natural Vector (ANV) method to address the problem of ambiguity in DNA sequences. We propose using ANV to predict the bases represented by non-standard codes in DNA sequences. Our approach involves developing a deep learning framework to establish a correspondence between DNA sequences (in FASTA format) and natural vectors, which encode relevant sequence properties. By training on a large dataset, we learn the distribution of these ambiguous base codes within the dataset. This method allows us to accurately predict masked or ambiguous bases in genomic fragments. It is particularly applicable to datasets, such as the COVID-19 genome data, which contain numerous non-standard base codes like R, Y, S, W, K, M, B, D, H, and V. By employing our algorithm, we can effectively estimate the corresponding standard bases and assign confidence scores to each prediction, aiding in the resolution of sequencing uncertainties. In addition, we will introduce research on gene regulatory relationships. Our ultimate goal:Given a genome sequence (1) Determine whether it is a regulatory factor (2) If so, which genomes does it have regulatory relationships with (3) Is this regulatory relationship promotion or inhibition.

北京雁栖湖应用数学研究院研究员
西安交通大学, 计算数学专业, 本科，硕士。
美国UIC大学，计算机科学，博士, 研究方向：非线性滤波控制，导师：Stephen S.-T Yau 丘成栋
毕业后主要在美国的无线通信领域工作。他曾在朗讯、阿尔卡特朗讯、诺基亚等公司担任高级工程师。
2024 年 1 月加入北京数学科学与应用研究院 （BIMSA），目前从事神经网络、人工智能、大数据、机器学习和生物数学方面的研究。