Search bioRxiv⌕ Search
Home/Archive/bioRxiv

bioRxiv · 10.1101/2022.11.03.515123

Machine learning-guided directed evolution for the development of small-molecule antibiotics originating from antimicrobial peptides

Zhang, H.·Wang, Y.·Huang, P.·Zhu, Y.·Li, X.·Chen, Z.·Liu, Y.·Jiang, J.·Gao, Y.·Huang, J.·Qin, Z.
View original source ↗Export metadata ↓

Abstract

Antimicrobial peptides (AMPs) are valuable alternatives to traditional antibiotics that possess a variety of potent biological activities by exerting immunomodulatory effects to clear difficult-to-treat infections. Understanding the structure-activity relationships (SARs) of AMPs can direct the synthesis of desirable therapeutics. In this study, we use machine learning-guided directed evolution to develop the lipopolysaccharide-binding domain (LBD), which acts as a functional domain of anti-lipopolysaccharide factor (ALF), a family of AMPs, identified from Marsupenaeus japonicus. We report the identification of LBDA-D as an output of this algorithm with the input of the original LBDMj sequence and show the NMR solution structure of LBDB, which possesses a circular extended structure with a disulfide crosslink in each terminus and two 310-helices and exhibits a broad antimicrobial spectrum. Scanning electron microscopy and transmission electron microscopy showed LBDB induced the formation of a cluster of bacteria wrapped in a flexible coating that ruptured and consequently killed the bacteria. The co-injection of LBDB and Vibrio alginolyticus, Staphylococcus aureus and another major pathogen in shrimp aquaculture white spot syndrome virus in vivo improved the survival of M. japonicus, indicating a promising therapeutic role of LBDB for infectious disease. The findings of this study pave the way for the rational drug design of activity-enhanced peptide antibiotics.

Explore related subjects

synthetic biologyPreprint

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Zhang, H., Wang, Y., Huang, P., Zhu, Y., Li, X., Chen, Z., Liu, Y., Jiang, J., Gao, Y., Huang, J., Qin, Z.. 2022-11-04. Machine learning-guided directed evolution for the development of small-molecule antibiotics originating from antimicrobial peptides. https://doi.org/10.1101/2022.11.03.515123

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Antibody design across observed and forecasted antigenic evolution

The therapeutic efficacy of antibodies is often limited by the evolution of their targets. Designing or eliciting antibodies that withstand such evolution is challenging as their binding interfaces must accommodate antigenic variation. Here, we introduce Germinal Multi, a computational framework for de novo antibody design and rescue that jointly optimizes a single paratope against multiple antigenic variants. We applied Germinal Multi to cancer and viral antigens, targeting both observed antigenic variation and prospective antibody escape. Multi-target optimization yielded antibodies that bound resistance-associated EGFR variants or that neutralized antigenically distinct historical SARS-CoV-2 variants, with recognition extending to variants withheld from the design process. Beyond acting on observed evolution, we redesigned the broadly neutralizing antibody BD55-1205 to restore neutralization of a prospective escape variant engineered with guidance from an evolutionary forecasting model, while preserving activity against the parental variant. Successful candidates were identified after screening tens of designs per campaign. Together, these results establish multi-target computational design as a strategy for generating antibody breadth across observed antigenic variation and demonstrate the potential of evolutionary forecasting to guide antibody redesign against prospective escape.

synthetic biology↗
bioRxiv2026

Chromosome-level genomics and cross-species enzyme screening enable high-titer bakuchiol production in Yarrowia lipolytica

The diversity of enzymes supporting bakuchiol biosynthesis remains incompletely characterized, limiting the repertoire available for microbial pathway engineering. Here, we combined chromosome-level genomics, cross-species enzyme screening and metabolic engineering to establish high-titer bakuchiol production in Yarrowia lipolytica. We assembled a 471.76-Mb Psoralea corylifolia genome anchored to 11 pseudochromosomes, with a contig N50 of 40.33 Mb. Genome-guided screening and functional validation identified four prenyltransferases from P. corylifolia and Vigna radiata whose N-terminally truncated variants supported bakuchiol production in yeast. These results extend experimental evidence for bakuchiol-producing activity across legume prenyltransferase sequences and species. For PcPT6, corresponding to the previously reported PcPT07, removing 40 N-terminal amino acids increased bakuchiol production by 53% relative to the full-length enzyme. We reconstructed bakuchiol biosynthesis in Y. lipolytica and reinforced aromatic precursor synthesis to increase p-coumaric acid supply. Coordinating mevalonate pathway reinforcement with increased t40PcPT6 gene dosage further improved bakuchiol production. Following fermentation optimization, titers reached 108.29 mg/L in shake flasks and 6.03 g/L in a 7.5-L fed-batch bioreactor. These findings connect an expanded set of functional prenyltransferases with coordinated pathway engineering, enabling gram-per-liter microbial bakuchiol production.

synthetic biology↗
bioRxiv2026

Associative Learning and Memory Erasure in a Synthetic Multicellular System

Learning is a basic form of adaptation that organisms display from neural systems to single cells. In associative learning, the pairing of two stimuli lets one of them evoke the response to the other. Engineering this behavior in a living cell is difficult, because coincidence detection, memory, and conditional readout must be wired into a single regulatory network. Here we build a consortium of five E. coli strains that distributes the operations of associative learning and communicates through one diffusible quorum-sensing signal. The same signal acts as the memory trace. One strain detects the coincidence of the two inputs, a second maintains the memory through positive feedback in the shared medium, a third reads it out on the conditioned stimulus, a fourth reports the unconditioned response, and a fifth erases the memory on command. We show experimentally that the assembled consortium learns the association, since a stimulus that is silent before training evokes a strong response after pairing. Memory is retained through a medium exchange in which both stimuli are removed, and then the conditioned stimulus alone evokes the response. The reporter is not produced without its input. An erase command returns the response to the background. A population model of the consortium accounts for the sequence of responses and gives design rules for writing, retention, and erasure, and a trade-off in storing memory in a shared extracellular signal. Distributing the operations over strains makes each of them separately tunable and testable and provides a route to microbial communities that learn from their history.

synthetic biology↗