评估调控基因组学预训练的DNA语言模型的表示能力
Ziqi Tang1, Nirali Somia1, Yiyang Yu2
1Simons Center for Quantitative Biology, Cold Spring Harbor Laboratory, Cold Spring Harbor, NY, USA.
Genome biology
|July 14, 2025
概括
基因组语言模型 (gLMs) 对于理解cis调节性DNA具有前景,但当前的预训练方法可能不会为功能基因组学任务提供比传统机器学习的优势.
科学领域:
- 基因组学就是基因组学.
- 计算生物学 计算生物学
- 机器学习 机器学习
背景情况:
- 基因组语言模型 (gLMs) 提供非编码基因组中cis-regulatory模式的无监督学习.
- 之前的研究表明,gLM在简单的调控基因组学任务上提高了预测性能.
- 确定gLM表示是否捕捉了基础的cis-regulatory生物学是一个开放的问题.
研究的目的:
- 评估预先训练有素的gLMs对预测和解释细胞类型特定的功能基因组学数据的表示能力.
- 评估六个主要的功能基因组学预测任务的DNA和RNA调节中的gLMs.
- 将gLM性能与传统机器学习方法进行比较.
主要方法:
- 预先训练的基因组语言模型的测试表示.
- 评估了六项功能基因组学预测任务的性能,这些任务涵盖了DNA和RNA调节.
- 将gLM性能与使用一热编码序列的传统机器学习模型进行比较.
主要成果:
- 测试当前预训练的gLM表示并没有比传统的机器学习带来实质性的优势.
- 从头开始训练的高度调整的监督模型在探索的数据集上表现优于预先训练的模型.
- 传统模型的性能与预先训练的gLM相匹敌或优于它们.
结论:
- 目前的gLM显示其有效捕获cis-regulatory生物学的能力存在显著差距.
- 对于非编码基因组的传统预训练策略可能需要重新评估.
- 需要进一步的研究来改善管理基因组学gLM预培训.
相关概念视频
02:02Cis-regulatory Sequences
10.1K
Cis-regulatory sequences are short fragments of non-coding DNA that are present on the same chromosomes as the genes that they regulate. These fragments serve as binding sites for transcriptional regulators, proteins that are responsible for controlling gene transcription and differential gene expression across cell types in eukaryotes. Cis-regulatory sequences can be close to the gene of interest or thousands of bases away in the DNA sequence; however, those sequences that are further away are...
10.1K
02:13Cooperative Binding of Transcription Regulators
6.6K
Transcriptional regulators bind to specific cis-regulatory sequences in the DNA to regulate gene transcription. These cis-regulatory sequences are very short, usually less than ten nucleotide pairs in length. The short length means that there is a high probability of the exact same sequence randomly occurring throughout the genome. Since regulators can also bind to groups of similar sequences, this further increases the chances of random binding. Transcriptional regulators form...
6.6K
01:37Epigenetic Regulation
3.1K
Epigenetic changes alter the physical structure of the DNA without changing the genetic sequence and often regulate whether genes are turned on or off. This regulation ensures that each cell produces only proteins necessary for its function. For example, proteins that promote bone growth are not produced in muscle cells. Epigenetic mechanisms play an essential role in healthy development. Conversely, precisely regulated epigenetic mechanisms are disrupted in diseases like cancer.
X-chromosome...
X-chromosome...
3.1K
01:11Genetic Lingo
104.9K
Overview
104.9K
01:23Regulation of Expression at Multiple Steps
1.0K
The gene expression in cells is regulated at different stages: (i) transcription, (ii) RNA processing, (iii) RNA localization, and (iv) translation. Transcriptional regulation is mediated by regulatory proteins such as transcription factors, activators, or repressors—these control gene expression by initiating or inhibiting the transcription of genes. Once a precursor or pre-mRNA is produced, it undergoes post-transcriptional modification, including 5' capping, splicing, and the...
1.0K
02:23Master Transcription Regulators
7.1K
Master transcription regulators are regulatory proteins that are predominantly responsible for regulating the expression of multiple genes. Often these genes work in concert to drive a complex process. Activation of a master transcription regulator can lead to a cascade of transcriptional activation necessary for that outcome. These regulators can directly bind to the regulatory sequences of the various genes involved, or they can indirectly regulate transcription by binding to regulatory...
7.1K