虚拟细胞需要背景,而不仅仅是规模
bioRxiv : the preprint server for biology
|February 23, 2026
概括
在生物学中,缩放人工智能 (AI) 模型不足以创建虚拟细胞. 主要的挑战是缺乏多样化的生物背景,而不是模型表达性,阻碍了准确的预测.
科学领域:
- 计算生物学 计算生物学
- 人工智能的人工智能
- 基因组学就是基因组学.
背景情况:
- 生物学中的AI领域正在迅速发展,旨在创建能够预测细胞反应的细胞 ('虚拟细胞') 的计算模型.
- 目前的方法集中在训练大型,高容量的模型,使用广泛的单细胞数据,灵感来自结构生物学和大型语言模型的成功.
研究的目的:
- 这篇论文认为,简单地扩展模型容量是解决虚拟细胞问题的不足.
- 确定的主要局限性是对各种生物背景的覆盖不足,而不是模型表达性的缺乏.
主要方法:
- 审查最近的研究,比较简单的基线与复杂的架构在特定的生物背景.
- 分析当前模型在不同生物环境中的概括能力.
- 将发现与有关可运输性的因果推理文献联系起来.
- 检查使用大规模 (2200万个细胞) 免疫学数据集的最先进模型.
主要成果:
- 简单的基线模型在一个单一的生物背景下进行评估时,可以与复杂的架构相提并论.
- 当前的模型在应用于新的或不同的生物环境时,表现不佳的概括性.
- 对大量免疫学数据集的分析揭示了当前最先进模型的局限性.
结论:
- 虚拟细胞问题基本上是一个"因果传输问题",它需要的不仅仅是从类似分布中增加数据.
- 未来的进步需要更多地强调语境多样性和因果表示学习.
- 这些方法应该补充,而不是取代,正在进行的努力,以缩放模型容量和数据量.
相关概念视频
01:22Cell Size
133.7K
Cell sizes vary widely among and within organisms. Bacterial cells range between 1-10 micrometers (μm)and are considerably smaller than most eukaryotic cells. The smallest bacteria are 0.1 μm in diameter—about a thousand times smaller than eukaryotic cells, which typically range from 10-100 μm.
Surface Area
Cells can take in nutrients and water via diffusion through the plasma membrane itself or through specific channels in the membrane. The area of the membrane surrounding...
Surface Area
Cells can take in nutrients and water via diffusion through the plasma membrane itself or through specific channels in the membrane. The area of the membrane surrounding...
133.7K
01:15What are Cells?
51.0K
Cells are the smallest and basic units of life, whether it is a single cell that forms the entire organism, e.g., in a bacterium, or trillions of them, e.g., in humans. No matter what organism a cell is a part of, they share specific characteristics.
Basic Characteristics of Cells
A living cell has a plasma membrane, a bilayer of lipids that separates the aqueous solution inside the cell called the cytoplasm from the outside environment.
Furthermore, a living cell possesses genetic information...
Basic Characteristics of Cells
A living cell has a plasma membrane, a bilayer of lipids that separates the aqueous solution inside the cell called the cytoplasm from the outside environment.
Furthermore, a living cell possesses genetic information...
51.0K
01:07What are Cells?
207.1K
Cells are the smallest and basic units of life, whether it is a single cell that forms the entire organism, e.g., in a bacterium or trillions of them, e.g., in humans. No matter what organism a cell is a part of, they share specific characteristics.
Basic Characteristics of Cells
A living cell has a plasma membrane, a bilayer of lipids that separates the aqueous solution inside the cell called the cytoplasm from the outside environment.
Furthermore, a living cell possesses genetic information...
Basic Characteristics of Cells
A living cell has a plasma membrane, a bilayer of lipids that separates the aqueous solution inside the cell called the cytoplasm from the outside environment.
Furthermore, a living cell possesses genetic information...
207.1K
01:13Cell Diversity
5.3K
The concept of a cell started with microscopic observations of dead cork tissue by Robert Hooke in 1665. Hooke coined the term "cell" based on the resemblance of the small subdivisions in the cork to the rooms that monks inhabited, called cells. About ten years later, Antonie van Leeuwenhoek became the first person to observe the living and moving cells under a microscope. In the century that followed, the theory that cells represented the basic unit of life developed.
Multicellular...
Multicellular...
5.3K
01:21Cell Culture
23.1K
Most vertebrate cells grow in vitro attached to a substrate as a monolayer, called adherent cultures. The flasks and plates used to grow cells are chemically treated to facilitate cell attachment. However, a few cell types, such as hematopoietic cells, can grow in a suspension. In contrast to adherent cultures, suspension cultures can grow in non-treated cultureware using magnetic stirrers or spinner flasks to agitate the culture media
23.1K
01:16Cell Lines
10.5K
A cell line is a population of cells grown in vitro that can be subcultured over several generations. Normal cells cease to divide after a certain number of cell divisions, a process known as replicative senescence. This number, called the Hayflick limit, was conceptualized by Leonard Hayflick in 1961 when he observed that fetal cells grown in culture could only divide 40-60 times. This limit is due to the shortening of the telomeres during each round of cell division, preventing cell division...
10.5K