当前,AI大模型逐渐被应用于蛋白质科学和生物信息学中,但其复杂性常常使人们无法解释神经网络如何从复杂的生物数据中提取和理解关键特征.为了理解这类计算模型如何拥有推断生物大分子的结构、功能和相互作用的能力,在前人关于预测治疗...当前,AI大模型逐渐被应用于蛋白质科学和生物信息学中,但其复杂性常常使人们无法解释神经网络如何从复杂的生物数据中提取和理解关键特征.为了理解这类计算模型如何拥有推断生物大分子的结构、功能和相互作用的能力,在前人关于预测治疗性抗体结合特异性的研究基础上进一步拓展,提出了基于通道注意力机制可解释的残差卷积神经网络.该网络能够有效预测具有不同氨基酸序列的抗体特异性结合概率,网络交叉验证的AUC(Area under Curve)达到0.943,与传统方法相比有显著提高.其次,通过非线性变换和积分梯度的方法获得各位点对于结合能力的贡献,从而推断出抗体序列的残基分布模式.提出的方法可以获得氨基酸序列背后潜在的信息,也能显著减小特异性抗体预测未知的突变空间,证明该网络不仅性能更优,对于理解复杂的神经网络背后的逻辑也有所帮助.展开更多
The dynamics of biomolecules span across a wide range of timescales,reflecting the complexity of free energy landscapes of biomolecules.Among these,the microsecond-tomillisecond(μs-ms)timescale dynamics are particula...The dynamics of biomolecules span across a wide range of timescales,reflecting the complexity of free energy landscapes of biomolecules.Among these,the microsecond-tomillisecond(μs-ms)timescale dynamics are particularly significant,offering detailed insights into the kinetic,thermodynamic,and structural aspects of biological function.Many critical biological processes,including enzyme catalysis,protein folding,ligand binding,and allosteric regulation,operate within this timescale.Nuclear magnetic resonance(NMR)spectroscopy is a powerful technique for probing molecular dynamics in this time window,commonly used NMR methods for investigatingμs-ms timescale dynamics include Carr-Purcell-Meiboom-Gill(CPMG)relaxation dispersion,chemical exchange saturation transfer(CEST),and rotating-frame longitudinal relaxation dispersion(R_(1ρ)relaxation dispersion).This review provides a brief ove rview of the fundamental principles and some recent advances of these methods,highlighting their interrelationships and applications in elucidating biomolecular dynamics.展开更多
文摘当前,AI大模型逐渐被应用于蛋白质科学和生物信息学中,但其复杂性常常使人们无法解释神经网络如何从复杂的生物数据中提取和理解关键特征.为了理解这类计算模型如何拥有推断生物大分子的结构、功能和相互作用的能力,在前人关于预测治疗性抗体结合特异性的研究基础上进一步拓展,提出了基于通道注意力机制可解释的残差卷积神经网络.该网络能够有效预测具有不同氨基酸序列的抗体特异性结合概率,网络交叉验证的AUC(Area under Curve)达到0.943,与传统方法相比有显著提高.其次,通过非线性变换和积分梯度的方法获得各位点对于结合能力的贡献,从而推断出抗体序列的残基分布模式.提出的方法可以获得氨基酸序列背后潜在的信息,也能显著减小特异性抗体预测未知的突变空间,证明该网络不仅性能更优,对于理解复杂的神经网络背后的逻辑也有所帮助.
基金financially supported by funds from the Natural Science Foundation of Beijing Municipality(Grant Number 7232251)the National Natural Science Foundation of China(Grant Number 22474006)。
文摘The dynamics of biomolecules span across a wide range of timescales,reflecting the complexity of free energy landscapes of biomolecules.Among these,the microsecond-tomillisecond(μs-ms)timescale dynamics are particularly significant,offering detailed insights into the kinetic,thermodynamic,and structural aspects of biological function.Many critical biological processes,including enzyme catalysis,protein folding,ligand binding,and allosteric regulation,operate within this timescale.Nuclear magnetic resonance(NMR)spectroscopy is a powerful technique for probing molecular dynamics in this time window,commonly used NMR methods for investigatingμs-ms timescale dynamics include Carr-Purcell-Meiboom-Gill(CPMG)relaxation dispersion,chemical exchange saturation transfer(CEST),and rotating-frame longitudinal relaxation dispersion(R_(1ρ)relaxation dispersion).This review provides a brief ove rview of the fundamental principles and some recent advances of these methods,highlighting their interrelationships and applications in elucidating biomolecular dynamics.
