Search bioRxiv⌕ Search
Home/Archive/bioRxiv

bioRxiv · 10.1101/2022.09.28.509940

Predicting enzyme substrate chemical structure with protein language models

Jinich, A.·Nazia, S. Z.·Tellez, A. V.·Rappoport, D.·Rhee, K. Y.
View original source ↗Export metadata ↓

Abstract

The number of unannotated or orphan enzymes vastly outnumber those for which the chemical structure of the substrates are known. While a number of enzyme function prediction algorithms exist, these often predict Enzyme Commission (EC) numbers or enzyme family, which limits their ability to generate experimentally testable hypotheses. Here, we harness protein language models, cheminformatics, and machine learning classification techniques to accelerate the annotation of orphan enzymes by predicting their substrates chemical structural class. We use the orphan enzymes of Mycobacterium tuberculosis as a case study, focusing on two protein families that are highly abundant in its proteome: the short-chain dehydrogenase/reductases (SDRs) and the S-adenosylmethionine (SAM)-dependent methyltransferases. Training machine learning classification models that take as input the protein sequence embeddings obtained from a pre-trained, self-supervised protein language model results in excellent accuracy for a wide variety of prediction tasks. These include redox cofactor preference for SDRs; small-molecule vs. polymer (i.e. protein, DNA or RNA) substrate preference for SAM-dependent methyltransferases; as well as more detailed chemical structural predictions for the preferred substrates of both enzyme families. We then use these trained classifiers to generate predictions for the full set of unannotated SDRs and SAM-methyltransferases in the proteomes of M. tuberculosis and other mycobacteria, generating a set of biochemically testable hypotheses. Our approach can be extended and generalized to other enzyme families and organisms, and we envision it will help accelerate the annotation of a large number of orphan enzymes. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=106 SRC="FIGDIR/small/509940v3_ufig1.gif" ALT="Figure 1"> View larger version (20K): org.highwire.dtl.DTLVardef@1dab89forg.highwire.dtl.DTLVardef@8eec65org.highwire.dtl.DTLVardef@141ff13org.highwire.dtl.DTLVardef@1d16212_HPS_FORMAT_FIGEXP M_FIG C_FIG

Explore related subjects

microbiologyPreprint

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Jinich, A., Nazia, S. Z., Tellez, A. V., Rappoport, D., Rhee, K. Y.. 2022-09-30. Predicting enzyme substrate chemical structure with protein language models. https://doi.org/10.1101/2022.09.28.509940

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

KEEP EXPLORING

Related preprints

bioRxiv2026

Prevalent Integration of both Xenotropic and Ecotropic Murine Leukemia Virus in the Human Genome of Xenografts

Humanized mouse models are widely used in medical research to represent human conditions (1, 2), but their fidelity could be compromised by mouse-originated infections. However, their susceptibility to mouse pathogens are rarely questioned (3). Here we screened 1,619 publicly available genomic sequencing runs from patient-derived xenograft (PDX) models for the integration of murine leukemia virus (MLV), a rodent retrovirus exists in the mouse genome as an endogenous retrovirus (4), within the human genome. We discovered substantial enrichment of human-MLV chimeric DNA in approximately 25% of them. Analyzing the human-MLV junctions, we found about 90% of them bearing the unmistakable hallmarks of retroviral integration, including the canonical MLV boundaries and the distinctive 4-bp target site duplications (5, 6). It was also assumed that mouse strains without the highly infectious Bxv1 provirus are safe from MLV contamination in xenografts (7, 8). However, we found many PDX models derived from Bxv1- strains, including the widely used NSG strain, are also contaminated by the supposedly rodent-specific ecotropic MLV (E-MLV). We further confirmed a complete E-MLV integrated into the human genome by both HiFi and sanger sequencing. Our discovery raises concerns that mouse-to-human retroviral transmission and integration is more common and less restrictive than previously appreciated. They support examination of MLV integration as a feature of humanized mouse models with implications on the fidelity of these models in recapitulating human cells and tissues. Our discovery also has underscored the importance of comprehensively inactivation of porcine endogenous retroviruses for the clinical application of xenotransplatation (9, 10).

microbiology↗
bioRxiv2026

Necroptosis is a targetable driver of severe coronavirus pathogenesis

Human-pathogenic coronaviruses differ dramatically in pathogenicity but the underlying mechanisms are incompletely understood. Here, we demonstrate that highly pathogenic coronaviruses trigger necroptosis in the infected cells and lung tissues, whereas necroptosis is not detected in cells infected by seasonal coronavirus. Mechanistically, highly pathogenic coronaviruses produce Z-RNA, activate ZBP1 and initiates necroptotic cell death signaling, triggering necroinflammation and lung injury. In contrast, seasonal coronaviruses produce little to no Z-RNA, and do not activate necroptosis. Importantly, small-molecule inhibition of necroptosis attenuates inflammatory cytokine expression and markedly ameliorates lung damage in mice infected with all three highly pathogenic coronaviruses, independent of virus clearance. Overall, our study identifies necroptosis as a dominant mediator of coronavirus pathogenesis, and positions necroptosis blockade as a new therapeutic target against severe coronavirus diseases.

microbiology↗
bioRxiv2026

POM121 regulates lentiviral nuclear import in a capsid-host factor-dependent manner

Lentiviruses such as human immunodeficiency virus 1 (HIV-1) access the nuclear lumen through the nuclear pore complex (NPC), a process coordinated by the interactions between the viral capsid and nucleoporins (NUPs). Here, using unbiased proteomic screens, we identified several capsid-associated NUPs, including POM121. Depletion of POM121 impaired HIV-1 infection in multiple cell lines and in primary CD4+ T cells, whereas equine infectious anaemia virus (EIAV) infection was enhanced, indicating that POM121 effect is virus-specific. A cytoplasmic carboxy-terminal fragment of POM121, containing its FG-rich region inhibited HIV-1 and a broader range of lentiviruses. Analysis of HIV-1 capsid mutants showed that physical interaction with POM121 does not predict functional dependence, as capsids retaining POM121 binding remained insensitive to its depletion. Strikingly, loss of cyclophilin A (CypA) completely abolished HIV-1 dependence on POM121, whereas depletion of CPSF6 produced only partial rescue, demonstrating that POM121 sensitivity is dictated by the broader network of capsid-host interactions. Finally, POM121 depletion impaired HIV-1 nuclear accumulation and selectively disrupted the nuclear import of other viral and cellular cargo. Together, these findings identify POM121 as a selective regulator of lentiviral nuclear entry whose functional contribution is determined by capsid and host-factor context.

microbiology↗