A Survey on Towards Genome Function Modeling from DNA Sequence and 3D Chromatin Structure

The Hong Kong University of Science and Technology (Guangzhou)

数据科学与分析学域

PhD Qualifying Examination

By Mr. LONG, Weicai

摘要

Regulatory genome function prediction aims to infer regulatory signals and spatial regulatory context from genomic data. This survey focuses on the question: given DNA sequences or 3D chromatin contact maps, how can computational models predict regulatory genome function? We first formulate this problem from two complementary views. The sequence view uses DNA sequence to predict regulatory readouts such as promoter and enhancer activity, splice-related signals, transcription factor binding, chromatin accessibility, histone modifications, and expression-related signals. The structure view uses chromatin contact maps to identify TADs, loops, chromatin domains, and spatial regulatory neighborhoods. Based on these two views, we review sequence-based methods, including supervised sequence-to-function models, self-supervised genome language models, and multimodal genome models. We then review structure-based methods, including heuristic TAD callers and learning-based structure models. Finally, we summarize major challenges and future directions, including sparse and redundant genomic signals, reliable evaluation beyond fixed benchmarks, raw-DNA understanding in general-purpose language models, limited task coverage in multimodal genome models, robust 3D regulatory neighborhood modeling, and cautious integration of sequence and 3D structure.

PQE Committee

• Chair: Prof. LUO, Qiong
• Prime Supervisor: Prof. ZHANG, Yanlin
• Co-Supervisor: Prof. WEI, Jiaheng
• Examiner: Prof. DING, Ningning