Automatic Cocrystal Detection by Raman Spectral Deconvolution-Based Novelty Analysis

Abstract

Cocrystals are important molecular adducts that have many advantages as a means of modifying the physicochemical properties of active pharmaceutical ingredients, including taste masking and improved solubility, bioavailability, and stability. As a result, the discovery of new cocrystals is of great interest to commercial drug discovery programs. Time-consuming manual analysis of the large volumes of data that emerge from large-scale cocrystal screening programs of up to 1000s of preparations poses a challenge. Raman spectroscopy has been shown to discriminate between cocrystals and physical mixtures and is easy to automate, allowing rapid screening of large numbers of potential cocrystals, but the spectral features that encode the information are often subtle (e.g., slight changes in peak positions or intensities). We have employed an automated signal processing routine based on a sparse decomposition algorithm to speed up the data processing steps while maintaining the accuracy of a trained spectroscopist. We used our algorithm to screen 31 potential cocrystal preparations and found that through the use of a computationally generated threshold, we could achieve a clear classification of cocrystals and physical mixtures in less than a minute, compared to several hours manually.

Mechanochemical Synthesis of Cocrystal: From Mechanism to Application

“…High-throughput screening appears to be a further aspect of concern. Unquestionably, more advanced equipment, such as innovative modified planetary mills, resonant acoustic hybrid porous plates (Figure i and j), and floating foam rack/sonic bath method, etc., is required, as are more effective and precise detecting methods. , Yamamoto et al also developed “cocktail cocrystal grinding” to grind the API with various CCFs with identical functional groups, which can increase the success rate by discovering partial structures of CCFs that form synthons with pharmaceuticals preferentially . Occasionally, the predictive capability of machine learning (Figure k) can also make cocrystal mechanochemical screening more purposeful and successful …”