预训练带有原子在分子量子属性的图形变压器,以改进ADMET建模
Alessio Fallani1,2, Ramil Nugmanov3, Jose Arjona-Medina2
1Department of Physics and Materials Science, University of Luxembourg, L-1511, Luxembourg , Luxembourg.
Journal of cheminformatics
|February 27, 2025
概括
使用原子级量子力学的预训图形变压器可以改善药物ADMET属性预测. 与其他预训练策略相比,这种方法在大型制药数据集上提供了更好的分子表示和性能.
科学领域:
- 计算化学和化学信息学
- 机器学习用于药物发现.
- 分子图表表示学习学习分子图表表示学习
背景情况:
- 准确预测吸收,分布,新陈代谢,分泌和毒性 (ADMET) 属性对于药物开发至关重要.
- 图形变压器架构显示出对分子性质预测的前景.
- 有效的预训练策略对于优化图形变压器在复杂的生物任务中的性能至关重要.
研究的目的:
- 评估不同预训练策略对用于ADMET属性预测的图形变压器模型的影响.
- 为了比较预训练与原子级量子力学特征,分子量子性质 (HOMO-LUMO差距) 和自我监督的原子掩饰.
- 分析学习的表示和它们与预测性能的相关性.
主要方法:
- 预训练使用三个不同的特征集的图形变压器架构:原子级量子力学,HOMO-LUMO间隙和原子掩饰.
- 在治疗数据共享ADMET数据集上微调预训练模型.
- 使用像注意力推广矩阵这样的技术分析潜在表示.
- 在公开的ADMET基准和用于微体清除的大规模内部数据集上评估模型性能.
主要成果:
- 用原子量子力学特性预训练的模型通常会产生优异的ADMET预测性能.
- 监督的预训练策略有效地保留了微调后的预训练信息.
- 原子量子力学预训练捕捉了低频拉普拉斯特征模式,并增强了原子环境表示.
- 大型药物数据集的性能在具有相似公共基准性能的模型之间可能有很大的差异,特别是在掩盖和原子级量子性质预训练方面.
结论:
- 预训练带有原子级量子力学特征的图形变压器是模拟类似药物化合物ADMET特性的一种高度有效的策略.
- 选择预训练显著影响学习的分子表示和下游任务性能.
- 分析潜在的表征和注意力机制为指导药物发现中的预训练策略选择提供了宝贵的见解.
相关概念视频
02:27Predicting Molecular Geometry
34.0K
VSEPR Theory for Determination of Electron Pair Geometries
34.0K
00:52¹H NMR of Conformationally Flexible Molecules: Temporal Resolution
796
At room temperature, the chair conformer of cyclohexane undergoes rapid ring flipping between two equivalent chair conformers at a rate of approximately 105 times per second. These two chair conformers are in equilibrium. The rapid ring flipping results in the interconversion of the axial proton to an equatorial proton and an equatorial to the axial proton. Such interconversions are too rapid and cannot be detected on the NMR timescale. Hence, the NMR spectrometer cannot distinguish between the...
796
01:25Atomic Absorption Spectroscopy: Atomization Methods
342
Atomic Absorption Spectroscopy (AAS) atomizes samples through flame atomization or electrothermal atomization. Flame atomization typically involves a nebulizer and spray chamber assembly to combine the sample with a fuel–oxidant mixture, creating a fine aerosol mist that enters a burner. Typically, the fuel and oxidant are combined in an approximately stoichiometric ratio. However, for atoms that are easily oxidized, a fuel-rich mixture may be more advantageous. Only about 5% of the...
342
03:24Hybridization of Atomic Orbitals I
46.2K
The mathematical expression known as the wave function, ψ, contains information about each orbital and the wavelike properties of electrons in an isolated atom. When atoms are bound together in a molecule, the wave functions combine to produce new mathematical descriptions that have different shapes. This process of combining the wave functions for atomic orbitals is called hybridization and is mathematically accomplished by the linear combination of atomic orbitals. The new orbitals that...
46.2K
02:45The Quantum-Mechanical Model of an Atom
41.8K
Shortly after de Broglie published his ideas that the electron in a hydrogen atom could be better thought of as being a circular standing wave instead of a particle moving in quantized circular orbits, Erwin Schrödinger extended de Broglie’s work by deriving what is now known as the Schrödinger equation. When Schrödinger applied his equation to hydrogen-like atoms, he was able to reproduce Bohr’s expression for the energy and, thus, the Rydberg formula governing hydrogen spectra.
41.8K
01:15¹H NMR of Conformationally Flexible Molecules: Variable-Temperature NMR
1.0K
The axial and equatorial protons in cyclohexane can be distinguished by performing a variable-temperature NMR experiment. In this process, except for one proton, the remaining eleven protons are replaced by deuterium. The deuterium substitution avoids the possible peak splitting caused by the spin-spin coupling between the adjacent protons. The remaining proton flips between the axial and equatorial positions.
1.0K