Tajs Patryk, Skarupski Mateusz, Rydzewski Jakub
Institute of Physics, Faculty of Physics, Astronomy and Informatics, Nicolaus Copernicus University, Grudziądzka 5, 87-100 Toruń, Poland.
J Chem Inf Model. 2025 Jul 28;65(14):7347-7351. doi: 10.1021/acs.jcim.5c01107. Epub 2025 Jul 14.
Unsupervised machine learning has recently gained much attention in the field of molecular dynamics (MD). Particularly, dimensionality reduction techniques have been regularly employed to analyze large volumes of high-dimensional MD data to gain insight into hidden information encoded in MD trajectories. Among many such techniques, t-distributed stochastic neighbor embedding (t-SNE) is especially popular. A parametric version of t-SNE that employs neural networks is less commonly known, yet it has demonstrated superior performance in dimensionality reduction compared to the standard implementation. Here, we present a Python package called NeuralTSNE with our implementation of parametric t-SNE. The implementation is done using the PyTorch library and the PyTorch Lightning framework and can be imported as a module or used from the command line. We show that NeuralTSNE offers an easy-to-use tool for the analysis of MD data.
无监督机器学习最近在分子动力学（MD）领域备受关注。特别是，降维技术经常被用于分析大量高维MD数据，以深入了解MD轨迹中编码的隐藏信息。在众多此类技术中，t分布随机邻域嵌入（t-SNE）尤为流行。一种采用神经网络的参数化t-SNE版本鲜为人知，但与标准实现相比，它在降维方面表现出了卓越的性能。在这里，我们展示了一个名为NeuralTSNE的Python包，其中包含我们对参数化t-SNE的实现。该实现是使用PyTorch库和PyTorch Lightning框架完成的，可以作为一个模块导入，也可以从命令行使用。我们表明，NeuralTSNE为MD数据分析提供了一个易于使用的工具。
