Enabling AI for glycoscience

‍

Background:

Artificial intelligence (AI) methods have been revolutionizing biology and medicine by enabling rapid analysis of large-scale biological data sets, accelerating the discovery of glycan-related biomarkers and therapeutic targets, and ultimately improving health outcomes. Despite the rapid progress in AI, “applications in glycocoscience are not yet widespread”, as explained by Daniel Bojar and Frederique Lisacek in their review paper entitled: "Glycoinformatics in the Artificial Intelligence Era".¹

The Challenge:

The integration of AI in glycoinformatics requires large, high-quality, glycan datasets to train models, and inconsistent or incorrect annotations can significantly affect the performance of these models. However, the isomeric complexity of glycans make it challenging to compile comprehensive datasets. Moreover, variations in experimental procedures and conditions in different laboratories can lead to discrepancies in the data, making it difficult to integrate and compare datasets.

Part of the problem is that the properties used to identify glycan isomers, such as chromatagaphic retention times or fragment intensities in a mass spectrometer, are extrinsic  properties of the molecule that depend upon the experimental conditions, which can vary from laboratory to laboratory.  In contrast, measuring an intrinsic property would provide transferable data that would be ideal for training machine-learning techniques.

‍

Isospec Analytics' Solution:

At Isospec Analytics we are revolutioning glycan analysis by measuring an infrared fingerprint of cryogenically cooled glycans inside of a mass spectrometer. Such a fingerprint is an intrinsic property that depends only on the quantum mechanical nature of the glycan molecule and is equisitly sensitive to its isomeric form. Measurement of such a spectrum is insensitive to experimental conditions, making it an ideal metric for transferable glycan databases to train AI models

‍

¹Bojar, D.; Lisacek, F. Glycoinformatics in the Artificial Intelligence Era. Chem. Rev. 2022, 122 (20), 15971-15988. DOI: 10.1021/acs.chemrev.2c00110.