UV-adVISor: Attention-Based Recurrent Neural Networks to Predict UV–Vis Spectra

Abstract

Ultraviolet-visible (UV-Vis) absorption spectra are routinely collected as part of highperformance liquid chromatography (HPLC) analysis systems and can be used to identify chemical reaction products by comparison to reference spectra. Here, we present UV-adVISor as a new computational tool for predicting UV-Vis spectra from a molecule's structure alone. UV-Vis prediction was approached as a sequence-to-sequence problem.We utilized Long-Short Term Memory and attention-based neural networks with Extended Connectivity Fingerprint diameter 6 or molecule SMILES to generate predictive models for UV-spectra. We have produced two spectrum datasets (Dataset I, N = 949 and Dataset II, N = 2222) using different compound collections and spectrum acquisition methods to train, validate, and test our models. We evaluated the prediction accuracy of the complete spectra by the correspondence of wavelengths of absorbance maxima and with a series of statistical measures (the best test set median model parameters are in parentheses for Model II), including RMSE (0.064), R 2 (0.71), and dynamic time warping (DTW, 0.194) of the entire spectrum curve. Scrambling molecule structures with experimental spectra during training resulted in a degraded R 2 , confirming the utility of the approaches for prediction. UV-adVISor is able to provide fast and accurate predictions for libraries of compounds.

Functional Output Regression for Machine Learning in Materials Science

“…Partitioning of the data set was performed according to the compound species to avoid multiple spectral profiles of the same compound leaking into the test set. In Urbina et al, 31 the results of applying two encoder−decoder architectures with LSTM cells and an attention mechanism, respectively, were reported, which were compared with the performance metrics of the present methods.…”

“…However, the methodology of supervised learning in such scenarios has not been well studied. For example, in the task of predicting the ultraviolet–visible (UV–vis) absorption spectra of molecules, the input variable is given by a vectorized molecular structure, and the output variable is given as a function defined on the domain of wavelengths that represents the optical absorbance . In the study of composite materials, it is important to qualitatively and quantitatively understand the influence of processing conditions such as temperature, pressure, and composition on the resulting microstructures.…”

“…To solve such problems, various types of deep generative models, such as the conditional GAN (cGAN) and encoder–decoder networks, can be applied. In fact, there have been several previous studies in which cGAN was applied to the prediction of microstructures, as described above, − and an encoder–decoder model was applied to predict the UV–vis absorption spectra of organic molecules . In addition, in statistical science, regression methods for functional output variables have long been studied in the context of functional data analysis, which may also be applicable to solving the aforementioned problems.…”

“…Comparison of the Prediction Accuracy for Optical Absorption Spectral Functions a Dataset I and Dataset II cover the UV−vis spectra, and the USGS spectral library covers data in the NIR domain. The performance metrics shown for the LSTM and attention-based encoder−decoder models are those reported in Urbina et al31 …”

Semi-automatic scheme for early-stage material search: developing solvent-solubility prediction of tetraphenylporphyrin derivatives securing chemical-space coverage

“…Second, the proposed method delivered adequate performance in the early stage of material search, but in future, the searching algorithm would be extended by combining with existing algorithms such as BOs. Moreover, the indicators FWHM and Area are regarded as approximate shape parameters of the UV-Vis absorption spectrum, and the prediction of the spectrum shape is still a challenging task [63]; thus, in future, we will continue to improve the prediction performance of the regression models. Despite these limitations, the proposed scheme demonstrated an effective example for material search projects.…”