Aug 2026· Journal of Agricultural and Food Chemistry· Vol 74 32, pp.
25452-25462
· 0 citations· 51 references
Medicine
TL;DR
A data-driven "digital twin" using deep learning-based Molecular Embeddings (MolE) and L1-regularized logistic regression to decode these structural determinants of enzymatic transglycosylation, providing a promising ligand-based virtual screening approach for identifying novel acceptors in enzymatic transglycosylation.
Abstract
Enzymatic transglycosylation by Deinococcus geothermalis amylosucrase (DgAS) offers a cost-effective route for enhancing the physicochemical properties of bioactive flavonoids. However, predicting DgAS acceptor specificity remains challenging due to complex structure-activity relationships. Here, we developed a data-driven "digital twin" using deep learning-based Molecular Embeddings (MolE) and L1-regularized logistic regression to decode these structural determinants. Training on experimentally validated substrates and physicochemical decoys, the MolE-based classifier achieved a 98.4% F1-score and 97.6% accuracy, outperforming traditional molecular fingerprints. SHAP analysis revealed a push-pull recognition mechanism, rewarding planar aglycone cores while penalizing steric constraints. Predictive power was validated through in vitro screening, identifying novel active acceptors (e.g., morin, 65.56% yield) and filtering out nonbinders like pinocembrin. Though constrained by substrate chemical stability, this acceptor-centric framework reduces trial-and-error, providing a promising ligand-based virtual screening approach for identifying novel acceptors in enzymatic transglycosylation.
Paracoccus denitrificans is a heterotrophic nitrifying–aerobic denitrifying bacterium widely used in wastewater treatment and as a model system for studying electron transport chains, making it a promising chassis for environmental synthetic biology. Precise translational control of gene expression is essential for engineering desired traits in this organism, yet the design of functional ribosome binding sites (RBS) in P. denitrificans remains hindered by limited understanding of their sequence–activity relationships. To address this gap, we systematically profiled 1335 native RBS sequences via integrated transcriptomic and proteomic analyses. We found that RBS strength is predominantly governed by a purine-rich Shine–Dalgarno motif located 5–8 bp upstream of the start codon, with specific A/G patterns in this region serving as key determinants. Building on this dataset, we developed and compared CNN, BiLSTM, and Transformer models for RBS strength prediction; among them, the CNN achieved the highest predictive correlation (Pearson r = 0.68). Additionally, a WGAN-GP framework was implemented to generate novel RBS sequences, which were subsequently filtered and evaluated by the prediction framework to enable the design of RBSs with user-specified strengths. Experimental validation showed a relatively strong correlation between predicted and measured strengths (Pearson r = 0.75). This work establishes the first deep learning-enabled RBS design platform for P. denitrificans, offering a robust tool for precise translational regulation and advancing synthetic biology applications in environmental biotechnology.
Zhiping Zheng, Shenghu Zhou, Yu Deng· Synthetic and Systems Biotec...· 0 citations
The solid-state fermentation of Jiangxiangxing Baijiu exhibits marked spatiotemporal heterogeneity in microbial communities and physicochemical parameters. We characterized the microbial succession and physicochemical dynamics of Zaopei (fermented grains) across Da-hui rounds (Rounds 3-5) and applied LimiX, a Transformer-based deep learning framework, for fermentation state monitoring. Spatial stratification explained 99% of microbial community variance (PERMANOVA, P < 0.001), with inner layers harboring greater bacterial diversity than surface layers. Staphylococcus and Weissella correlated negatively with starch and reducing sugar, while Oceanobacillus and Acetobacter strongly discriminated among rounds. LimiX achieved an AUC of 1.000 for round classification and outperformed GLM and Random Forest (RF) in discriminating fermentation layers. External validation yielded 100% accuracy for round differentiation and 92.86% for temporal stage prediction. SHAP analysis identified Klebsiella and Thermoascus as the primary drivers of spatial and temporal stratification, respectively. These results demonstrate the capability of deep learning to decode complex fermentation data and provide a theoretical and technical basis for the intelligent digitalization of Baijiu production. PRACTICAL APPLICATIONS: A Transformer-based deep learning model was developed to integrate amplicon sequencing data and physicochemical indices for decoding the spatiotemporal dynamics of fermented grains in Jiangxiangxing Baijiu fermentation. This framework supports intelligent process monitoring and quality control, with potential applicability to other complex food fermentation systems.
Yingyu Huo, Fei Zhang, Gang Yang et al.· Journal of Food Science· 0 citations
A transfer learning-based predictor for acidophilic and alkalophilic proteins, trained on a curated non-redundant dataset, and an integrated pipeline for mining and engineering alkalophilic and thermophilic enzymes, combining sequence-based prediction, generative modeling, and multi-parameter virtual screening are developed.
Ruohan Zhang, Yiyang Zhang, Zhonghao Deng et al.· Bioresource Technology· 0 citations
TRACER (terpene rearrangement annotation via co-attentive enzyme-product representation), a multimodal framework mapping the latent associations between sequence-derived enzyme representations and product chemotypes, establishes a predictive paradigm for the rational discovery and mechanistic elucidation of complex terpene architectures.
The growing demand for sustainable, clean-label food ingredients has driven the search for alternative sources of functional biopolymers like pectin. This study explores fruit peel waste from Neolamarckia cadamba, an underutilized agro-resource, as a novel substrate for pectin extraction. The extraction process was optimized using Response Surface Methodology (RSM), and the predictive capability of the model was further enhanced through Artificial Neural Network (ANN) analysis. ANN outperformed RSM in all statistical metrics (R2, RMSE, MAE), better capturing non-linear process dynamics. Under optimal conditions (pH 2.05, 90 °C, 60 min), a maximum pectin yield of 15.89% (w/w) was achieved. Structural characterization using Fourier Transform Infrared Spectroscopy (FTIR), Liquid Chromatography – Mass Spectrometry (LC-MS), and 1H Nuclear Magnetic Resonance (1H NMR) confirmed the presence of galacturonic acid-rich homogalacturonan. The extracted pectin was subsequently evaluated as a gelling agent in mixed fruit jam formulations at concentrations of 2%, 4%, and 6%, and its performance was compared with that of a commercial jam. The 4% formulation showed favourable texture – hardness, chewiness, and springiness, closely matching the commercial product. Nutritional analysis revealed lower sugar (40.64–42.65 g/100 g), higher protein (4.2–4.5 g/100 g), and enhanced mineral content in the experimental jams. This work highlights N. cadamba peel as a promising, sustainable pectin source and showcases the strength of RSM–ANN integration for optimizing biopolymer extraction.
N. S. S. Kumar, Sandeep Singh Rana, N. M. Sarbon et al.· RSC Advances· 0 citations
Polyethylene terephthalate (PET) hydrolase-based biodegradation offers a promising route for plastic waste remediation, yet the dynamic origins of high activity and their linkage across catalytic stages still require further elucidation. Here, we focus on the Leaf-Branch Compost Cutinase (LCC) system to reveal the possible mechanistic origins underlying the elevated activity of the LCC variants. By integrating molecular dynamics simulations, enhanced sampling, and deep learning-assisted network analysis, we systematically investigate the critical prereactive stage of amorphous PET adsorption and substrate binding. We identify three mechanistic origins underlying the high activity of LCC-LANL in the prereactive state and clarify its structure–dynamics–function relationship: (i) enhanced interaction strength coupled with an active-pocket orientation that, despite not directly facing the amorphous PET surface, maintains closer proximity to it than LCC-WT, thus promoting substrate recruitment; (ii) higher occupancy of the PET ester bond near the catalytic triad, which forms the basis for catalysis, coupled with the efficient dynamic interchange between the “W” and coiled conformations near the catalytic triad; and (iii) the enhancement of prereactive organization through long-range allosteric communication by distal mutations in LCC-LANL. Additionally, we propose a region-specific cooperative optimization strategy tailored to domain-specific functional roles and distill six design principles for efficient PETases. In summary, this work elucidates the prereactive origins underlying the high activity of LCC-LANL, paving the way for future studies on actual catalytic PET hydrolysis.
Jiawen Wang, Haozhe Pan, Huilong Dong et al.· Journal of Chemical Informat...· 0 citations