Mapping the glycosyltransferase fold landscape using interpretable deep learning

Abstract

Glycosyltransferases (GTs) play fundamental roles in nearly all cellular processes through the biosynthesis of complex carbohydrates and glycosylation of diverse protein and small molecule substrates. The extensive structural and functional diversification of GTs presents a major challenge in mapping the relationships connecting sequence, structure, fold and function using traditional bioinformatics approaches. Here, we present a convolutional neural network with attention (CNN-attention) based deep learning model that leverages simple secondary structure representations generated from primary sequences to provide GT fold prediction with high accuracy. The model learns distinguishing secondary structure features free of primary sequence alignment constraints and is highly interpretable. It delineates sequence and structural features characteristic of individual fold types, while classifying them into distinct clusters that group evolutionarily divergent families based on shared secondary structural features. We further extend our model to classify GT families of unknown folds and variants of known folds. By identifying families that are likely to adopt novel folds such as GT91, GT96 and GT97, our studies expand the GT fold landscape and prioritize targets for future structural studies.

Structural basis for the synthesis of the core 1 structure by C1GalT1

“…Hereafter we will discuss only molecule A because it contains a better-defined density for the ligands. In addition, Dm C1GalT1 also contained the four conserved landmark features among GT-A GTs 42 : the DxD motif for metal cation interactions (Asp181-X-Asp183), a “glycine-rich” loop facing the acceptor and donor sugar site located in Dm C1GalT1 at loop β5-β6, an “xED” motif at the beginning of α6 in Dm C1GalT1 harboring the catalytic base (Asp255, see further experiments below), and a “C-His” residue that coordinates with the metal ion (His324) (Fig. 3b and Supplementary Fig.…”

De Novo Peptide and Protein Design Using Generative Adversarial Networks: An Update

“…Moreover, this proposed model can produce interpretable activation maps, with the help of attention modules, to provide meaningful biological inferences, where the activation maps can maintain spatial information (i.e., features) to determine the key protein sequence regions. 119 It is apparent that more interpretable GAN-based algorithms are warranted, and hence, we could identify explainable features pinpointed by the GAN model.…”

“…17 It is also worth mentioning that not only the GAN model but also other deep learning algorithms (e.g., convolutional neural network modules) have this issue. 17 For instance, Taujale et al 119 proposed an interpretable deep learning model to predict protein folds for glycosyltransferase protein families, where the model was based on a convolutional neural network module with the addition of attention modules. Moreover, this proposed model can produce interpretable activation maps, with the help of attention modules, to provide meaningful biological inferences, where the activation maps can maintain spatial information (i.e., features) to determine the key protein sequence regions.…”

Glycosyltransferases: Mining, engineering and applications in biosynthesis of glycosylated plant natural products

“…In addition, the development of artificial intelligence provides a unique avenue for the identification of target enzymes through rapidly expanding sequence databases. Recent advances in deep learning models for feature extraction and patter recognition for sequence classification and functional prediction of enzymes in large datasets have also facilitated the discovery of novel GTs [ 124 , 125 ]. Yang et al.…”