Biology subjects

Geng, P. X.

Publications and source records attributed to Geng, P. X..

5 recordsLinked to original sources

bioRxiv2026

eBiota: Designing microbial communities from large seed pools with desired function using rapid optimization and deep learning

Designing microbial communities to generate target products is crucial for biotechnology, agriculture, and disease treatment. However, rationally designing such communities from large seed pools has become a major challenge, as the rapidly expanding number of complete microbial genomes greatly expands the search space and sharply increases the required screening time and computational cost. Here, we introduce eBiota, a platform for ab initio design of microbial communities from a pool of 21,514 strains to generate target products. eBiota not only identifies optimal strain combinations but also simulates community behaviors, including microbial interactions and relative abundances. eBiota integrates three modules: CoreBFS, a graph-based search algorithm that rapidly screens for bacteria with complete metabolic pathways related to the target product; ProdFBA, an extended flux balance analysis that identifies microbial consortia with maximal production efficiency; and DeepCooc, a deep learning model trained on 23,323 microbiome samples across various environments to infer co-occurrence patterns. We validated eBiotas capabilities in microbial community design and production efficiency calculation using public microbiome datasets, ranging from single strains to six-member consortia. Further in vitro experiments involving 94 strains confirmed eBiotas ability to identify species that inhibit pathogen growth and to accurately model the relative abundances within complex microbial communities. As an initial digital twin, eBiota provides a powerful platform for the rational design of functional microbial communities, offering new opportunities for metabolic engineering and synthetic biology.

bioengineering↗

bioRxiv2026

KuafuPrimer: Machine learning empowers the design of 16S amplicon sequencing primers toward minimal bias for bacterial communities

Amplicon sequencing protocol targeting the 16S rRNA gene is a widely used and cost-effective method for exploring bacterial communities. However, its performance is often limited by primer bias arising from the arbitrary use of universal primers across diverse microbial communities and habitats. We propose KuafuPrimer to design the optimal 16S rRNA gene primers toward minimal bias for targeted bacterial communities, using few-shot machine learning to guide the primer design procedure based on a small number of samples. Simulations on 809 samples across 26 representative environments and habitats showed that KuafuPrimer-designed primers outperformed the universal primers in taxonomic accuracy, achieving an averaged 16.31% relative reduction in primer bias, with reductions up to 46.08% in plant samples. Notably, KuafuPrimer detected 29 rare and key taxa undetectable by the universal primers. Validation with 317 longitudinal gut microbiota samples demonstrated that KuafuPrimer-designed primers consistently outperformed the universal primers across temporal, individual, and cohort levels, with relative bias reductions of 5.03%, 3.53%, and 3.10%, respectively. Finally, in real PCR experiments on human gut samples from Clostridioides difficile-infected and healthy groups showed that polymerase chain reaction products using KuafuPrimer-designed primers correlated better with metagenomic data compared to the universal primers. More importantly, KuafuPrimer successfully detected Clostridioides difficile, the key pathogen missed by the universal primers, highlighting its potential for improving clinical diagnostics. In summary, KuafuPrimer provides a machine learning-based primer design strategy for targeted bacterial communities, with demonstrated utility in large-scale microbiome initiatives, longitudinal surveys and clinical diagnostics.

bioinformatics↗

bioRxiv2025

Tripleknock: predicting lethal effect of three-gene knockout in bacteria by deep learning

Investigating the lethal effect of multi-gene knockout is essential for discovering novel antibiotics targets and metabolic engineering. Unlike single genes or gene pairs, three-gene combinations involve more intricate interactions, making experimental screening time-consuming. Computational methods, particularly Genome-scale metabolic Model (GEM)-based Flux Balance Analysis (FBA), requires constructing new GEMs from experimental data, limiting its use for new species. Moreover, using FBA for three-gene knockout screening could take several years. Therefore, a faster and GEMs-independent approach is needed to facilitate genome-wide three-gene knockout screening. Here, we introduce Tripleknock, for predicting the lethal effects of three-gene knockouts. Tripleknock was trained using whole-genome data from Escherichia coli K-12 MG1655, and three-gene knockout simulations using FBA. The model uses a threshold of 90% reduction in cell growth to define lethal effect as the prediction output. Compared to FBA, Tripleknock achieves predictions approximately 20 times faster, reaching an average cross-species F1 score of 0.77 on six pathogenic species within the Enterobacteriaceae family. For closely related species such as pathogenic E. coli and Shigella, Tripleknock reaches F1 scores exceeding 0.83. To our knowledge, Tripleknock is the first end-to-end model for predicting lethal effects of three-gene knockout in bacteria. Data availabilityTripleknock is publicly available at: https://github.com/Peneapple/Tripleknock

bioinformatics↗

bioRxiv2025

Developing Foundation Models for Predicting Viral Animal Host Range in Intelligent Surveillance

Emerging human infectious viruses originating from animals continue to pose a persistent threat to global public health. Understanding the host range of animal viruses is crucial for identifying potential spillover pathways and mitigating the risk of future pandemics. Here, we present VirHRanger, a prediction method that integrates foundation models trained on viral genome and protein sequences, alongside genomic and protein compositional traits, viral phylogeny, and protein-protein interactions. To systematically predict the animal host range, VirHRanger incorporates host taxonomy-aware neural networks trained on a comprehensive collection of animal-virus associations spanning mammals, birds, and arthropods. Within a dataset of 4,006 virus species spanning 99 viral families, our model achieved robust performance with a micro-averaged AUROC of 0.938 across all host categories, demonstrating its effectiveness in capturing generalizable host signals from viral genetic data. On a dataset of 315 novel viruses, which are associated with key reservoir animal hosts and insect vectors, VirHRanger notably outperformed the homology-based method, exhibiting a strong generalizability to novel viruses. Furthermore, VirHRanger identified host range variations among closely related viruses within the Coronaviridae family and successfully predicted the ability of SARS-CoV-2 to infect humans and other animal hosts. These findings highlight the potential of VirHRanger to transform sequencing data into timely insights for disease control during the early stages of zoonotic outbreaks.

bioinformatics↗

bioRxiv2024

Characterizing the Binding Mechanism of Sparsentan to the Type-2 Angiotensin II Receptor Using AutoDock Vina

The angiotensin II type-2 receptor (AT2R) is known to have a significant impact on cardiovascular physiology, and the elucidation of its drug interactions is crucial for advancing therapeutic interventions. Sparsentan is an antagonist of angiotensin II, which has received FDA approval due to its demonstrated efficacy in ameliorating proteinuria. While sparsentan exhibits a preferential affinity for angiotensin II type-1 receptor (AT1R) and may primarily target it, DrugBank identifies it as a drug that targets AT2R. Present databases lack detailed structural data on the interaction mechanism between sparsentan and AT2R, leaving no definitive evidence to confirm or refute an interaction with AT2R. This study aimed to explore the potential for an interaction between sparsentan and AT2R. Molecular docking simulations were conducted using AutoDock Vina to predict the binding conformation. The docking parameters were meticulously optimized to ensure the accuracy and reliability of the simulation results. Subsequently, the most stable sparsentan-AT2R complex was determined. We identified four primary types of interactions at the binding site containing hydrogen bonds, hydrophobic interactions, {pi}-cation interactions, and {pi}-stacking, that are likely to contribute to the affinity of sparsentan for AT2R. Our study indicates that sparsentan may potentially interact with AT2R, our results may serve as a valuable reference for elucidating the mechanism of sparsentan action and for guiding the design of AT2R-targeted drugs.

biophysics↗