Search bioRxiv⌕ Search

bioRxiv · 10.64898/2026.07.20.739108

Combining Stability-Centered Atomistic Design with Machine Learning for Targeted Enzyme Optimization

Abstract

FuncLib and high-throughput FuncLib (htFuncLib) generate diverse, functional protein libraries using a stability-centered design; however, this substrate-independent approach lacks target-specific functional constraints. We developed a machine-learning-assisted enzyme-engineering (MLEE) workflow that adds substrate-specific functional information to htFuncLib through an initial screening and sequencing round. The system was benchmarked using previously published four-position fitness landscapes of three different proteins. The MLEE workflow successfully generated compact libraries enriched in globally high-fitness variants. After the initial training phase, an MLEE-enriched library of just 12 variants increased the hit rate for the global top-0.05% variants by 5- to 12-fold relative to the htFuncLib baseline. Screening a larger set of 96 variants recovered at least one of these top-performing enzymes in 61.3-99.4% of the simulations. We then applied MLEE to MthUPO-catalyzed {beta}-damascone hydroxylation. Across two rounds, 506 distinct variants were screened and sequenced. While the initial substrate-independent htFuncLib library yielded 14% of variants with activity above the wild type, the MLEE-enriched library increased this hit rate to 90% (97 of 108 variants) with activity above the wild type. The best variant increased the turnover number for 4-hydroxy-{beta}-damascone by 11.8-fold and achieved >99% regioisomeric excess. MLEE may bypass the need for transition-state models and reduce the effort required for obtaining high-activity variants. TABLE OF CONTENT O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=100 SRC="FIGDIR/small/739108v1_ufig1.gif" ALT="Figure 1"> View larger version (18K): org.highwire.dtl.DTLVardef@1501d4aorg.highwire.dtl.DTLVardef@b32171org.highwire.dtl.DTLVardef@58b249org.highwire.dtl.DTLVardef@120f8a6_HPS_FORMAT_FIGEXP M_FIG C_FIG

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Wan, L., Bagherpoor Helabad, M., Fraedrich, L., Fleishman, S. J., Weissenborn, M.. 2026-07-21. Combining Stability-Centered Atomistic Design with Machine Learning for Targeted Enzyme Optimization. https://doi.org/10.64898/2026.07.20.739108

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

KEEP EXPLORING

Related preprints

Deep reinforcement learning-driven discovery of a MsbA-targeted small-molecule antibiotic for the treatment of Acinetobacter baumannii infection

Antibiotics with new mechanisms are highly pursued to address the threat of infections caused by drug-resistant Gram-negative bacteria. Targeting MsbA, a key protein of the lipopolysaccharide biosynthesis pathway, represents a promising strategy to discover new classes of antibiotics. However, currently available MsbA-targeted molecules either lack sufficient potency or have unfavorable properties, necessitating expansion of chemical space. In this study, we chose the most promising cerastecin Cpd 4 as the template, and used two Artificial Intelligence (AI)-based tools, i.e. Link-INVENT and AutoMolDesigner for molecular design, performed chemical derivatization and antibacterial activity evaluation, which led to the discovery of Y-11 (MIC for A. baumannii: 0.5 g/mL). Encouragingly, Y-11 showed equivalent potency to Cpd4 for carbapenem-resistant A. baumannii, and less cytotoxicity and hemolysis as well as lower spontaneous resistance frequency. In vivo efficacy study demonstrated that Y-11 could effectively reduce bacterial loads in the mice infected by A. baumannii. The following mechanism study including molecular dynamics simulation, biochemical assay, and transmission electron microscope (TEM) analysis suggested that Y-11 inhibited the transport of lipooligosaccharide and impaired the formation of outer membrane, probably by competitively binding to the substrate binding site of MsbA and modulating ATPase activity. Taken together, we have discovered a MsbA-targeted small molecule Y-11 via AI-driven drug design, which provides a foundation for future antibiotic development.

biochemistry↗

Dynamic architecture of the Rixosome reveals mechanism of activation and ITS2 processing

Eukaryotic ribosome assembly requires the coordinated processing and extensive remodeling of pre-rRNAs. During late nuclear maturation of the 60S subunit, sequential removal of the internal transcribed spacer 2 (ITS2) is initiated by endonucleolytic cleavage at site C2 by the conserved Las1 nuclease. Las1 acts together with the kinase Grc3 and the Rix1 complex to form the Rixosome, which also functions in transcriptional regulation. However, the assembly of the Rixosome, its recruitment to pre-ribosomes, and its activation for ITS2 cleavage remain unclear. Here, we present cryo-EM structures of the human LAS1 complex, two structures of the isolated Rixosome and nine transition states of Rix1-bound pre-60S particles from Schizosaccharomyces pombe. These structures reveal a dynamic Rixosome architecture in which the heterotetrameric Las1 complex engages one or two copies of the Rix1 complex. Rix1 binding is highly flexible in the human Rixosome but rigid in the yeast complex. The isolated yeast Rixosome remains inactive, but binding to the pre-60S particle triggers a structural rearrangement that allows for substrate engagement and activation of the nuclease. Together, our results define the dynamic architecture of the Rixosome and provide a structural framework for ITS2 processing during nuclear maturation of the eukaryotic 60S ribosomal subunit.

biochemistry↗

SGFP-Grid Split GFP Graphene Grids

Affinity graphene grids provide a promising approach for selective protein capture in cryo-EM. Here, we introduce a split-GFP graphene grid platform(SGFP-G), in which graphene-conjugated GFP 1-10 selectively captures GFP11 tagged proteins from low concentration samples or cell lysates. This platform enables rapid assessment of target protein enrichment and particle distribution before vitrification via fluorescence imaging, while the grid design positions captured proteins away from the graphene surface and air-water interface. We also introduce a unique strategy to minimize nonspecific protein adsorption, thereby improving the selective enrichment of target proteins on this grid. Using GFP11-tagged apoferritin, we demonstrate fluorescence guided protein capture and obtain a 2.58 [A] cryoEM reconstruction, establishing SGFP-G as an affinity grid platform for high resolution structural studies with reduced sample requirements.

biochemistry↗