改进亚特拉斯规模的单细胞注释模型,具有层次的交叉损失
Sebastiano Cultrera di Montesano1, Davide D'Ascenzo2,3, Srivatsan Raghavan4,5,6,7
1Broad Institute of MIT and Harvard, Cambridge, MA, USA. scultrer@broadinstitute.org.
Nature computational science
|January 30, 2026
概括
我们开发了一个层次的交叉损失,以改善单细胞RNA测序 (scRNA-seq) 数据中的细胞类型注释. 这种方法在不增加计算成本的情况下提高了对新数据的模型性能,强调了数据生成对算法概括性的重要性.
科学领域:
- 计算生物学是一种计算生物学.
- 基因组学就是基因组学.
- 生物信息学是一种生物信息学.
背景情况:
- 准确的细胞类型注释对于单细胞RNA测序 (scRNA-seq) 数据分析至关重要.
- 现有的计算模型往往无法利用细胞类型本体学的固有层次结构.
- 这种限制阻碍了预测模型的通用性.
研究的目的:
- 引入一种新的损失函数,将生物层次纳入用于细胞类型注释的机器学习模型中.
- 为了提高计算模型在分布外的scRNA-seq数据集上的性能.
- 引导未来的研究向数据生成策略指导,以提高算法概括性.
主要方法:
- 开发了一个分层的交叉损失函数.
- 将损失函数应用于各种机器学习架构,包括线性模型和变压器.
- 对分布外数据集的模型性能进行评估,以评估可通用性.
主要成果:
- 在不同的模型架构中,等级交叉损失提高了12-15%的分布外性能.
- 该修改没有引入额外的计算成本.
- 在不增加模型复杂性的情况下实现了性能增长.
结论:
- 将生物层次纳入模型培训目标是改善细胞类型注释的有效策略.
- 专注于生成加强注释细胞类型之间的连接的新数据,对于开发更可概括的算法至关重要.
- 这种方法为推进scRNA-seq分析提供了一个更有前途的途径,而不仅仅是增加模型复杂性.
相关概念视频
Entropy
36.0K
Salt particles that have dissolved in water never spontaneously come back together in solution to reform solid particles. Moreover, a gas that has expanded in a vacuum remains dispersed and never spontaneously reassembles. The unidirectional nature of these phenomena is the result of a thermodynamic state function called entropy (S). Entropy is the measure of the extent to which the energy is dispersed throughout a system, or in other words, it is proportional to the degree of disorder of a...
36.0K
Entropy
3.6K
The first law of thermodynamics is quantitatively formulated via an equation relating the internal energy of a system, the heat exchanged by it, and the work done on it. A quantitative formulation of the second law of thermodynamics leads to defining a state function, the entropy.
When an ideal gas expands isothermally, the disorder in the gas increases. From the molecular perspective, the gas molecules have more volume to move around in.
Consider an infinitesimal step in the expansion, which...
When an ideal gas expands isothermally, the disorder in the gas increases. From the molecular perspective, the gas molecules have more volume to move around in.
Consider an infinitesimal step in the expansion, which...
3.6K
Entropy within the Cell
12.9K
A living cell's primary tasks of obtaining, transforming, and using energy to do work may seem simple. However, the second law of thermodynamics explains why these tasks are harder than they appear. None of the energy transfers in the universe are completely efficient. In every energy transfer, some amount of energy is lost in a form that is unusable. In most cases, this form is heat energy. Thermodynamically, heat energy is defined as the energy transferred from one system to another that...
12.9K
Standard Entropy Change for a Reaction
24.4K
Entropy is a state function, so the standard entropy change for a chemical reaction (ΔS°rxn) can be calculated from the difference in standard entropy between the products and the reactants.
24.4K
Crossing Over
171.9K
Unlike mitosis, meiosis aims for genetic diversity in its creation of haploid gametes. Dividing germ cells first begin this process in prophase I, where each chromosome—replicated in S phase—is now composed of two sister chromatids (identical copies) joined centrally.
The homologous pairs of sister chromosomes—one from the maternal and one from the paternal genome—then begin to align alongside each other lengthwise, matching corresponding DNA positions in a process...
The homologous pairs of sister chromosomes—one from the maternal and one from the paternal genome—then begin to align alongside each other lengthwise, matching corresponding DNA positions in a process...
171.9K
Genome Annotation and Assembly
21.0K
The genome refers to all of the genetic material in an organism. It can range from a few million base pairs in microbial cells to several billion base pairs in many eukaryotic organisms. Genome assembly refers to the process of taking the DNA sequencing data and putting it all back together in a correct order to create a close representation of the original genome. This is followed by the identification of functional elements on the newly assembled genome, a process called genome annotation.
21.0K


