Biology subjects

Wilke, C. O.

Publications and source records attributed to Wilke, C. O..

At least 19 recordsLinked to original sources

bioRxiv2026

A structure-guided classification framework reveals the diversity and catalytic architecture of BECR ribonuclease

Microorganisms across all domains of life engage in molecular conflict, deploying toxins to inhibit competitors or respond to biological threats. Among these, ribonuclease toxins are particularly widespread and diverse. A substantial fraction is associated with the BECR fold, a compact /{beta} architecture that supports RNase activity despite extensive divergence. Although several canonical members are well characterized, many BECR-fold proteins remain difficult to identify because of low sequence similarity, variation in catalytic residues, and structural elaborations that obscure evolutionary relationships. The growing availability of high-confidence protein structure predictions provides an opportunity to reassess this deeply divergent protein landscape. Here, we integrate iterative profile-HMM searches, profile-similarity networks, structural analyses, active-site mapping, and genomic context to examine BECR proteins across the tree of life. Our analysis resolves an expanded BECR-fold landscape comprising canonical BECR and BECR-like superfamilies, refines the organization of canonical BECR proteins and identifies previously unrecognized families. We further validate BECR-Tox2 as a toxin neutralized by a cognate immunity protein and show that its homologs occur in both Menshen-like anti-phage systems and polymorphic toxin loci. Together, these findings expand and clarify the BECR-fold landscape and provide a framework for identifying and interpreting highly divergent proteins of this fold.

microbiology↗

bioRxiv2026

Overestimating zero-shot fitness prediction: Broad benchmarks mask local failures and practical limitations

Deep learning models have emerged as promising tools in protein engineering. In particular, they can be used to predict mutation fitness without the need for task-specific training, a process known as zero-shot prediction. However, the respective strengths and limitations of zero-shot predictions remain poorly understood. Here, we argue that commonly used large-scale benchmarks obscure important failure modes relevant to practical protein engineering, including an inability to pinpoint highly fit mutations or variants driving new-to-nature functions. Beyond these practical failures, we identify a fundamental limitation of zero-shot prediction: a generic fitness score cannot simultaneously optimize for distinct, competing engineering targets, meaning it is inherently disconnected from the phenotype of interest. Moreover, in a systematic comparison of a wide range of available models, we demonstrate that most models show comparable zero-shot performance, irrespective of model architecture and/or input modality (sequence vs. structure). Ultimately, we find that zero-shot predictions serve only as coarse filters separating fit mutations from deleterious ones, failing to reliably identify the mutations that would be most valuable in protein engineering.

bioengineering↗

bioRxiv2026

Intrinsic dataset features drive mutational effect prediction by protein language models

Protein language models (pLMs) are commonly used for predicting protein fitness landscapes, but their wide range of performance across datasets remains poorly understood. We evaluated supervised transfer learning on 41 viral and 33 cellular deep-mutational-scanning (DMS) datasets using embeddings from multiple pLMs. We observed consistently lower predictive performance on viral datasets compared to cellular datasets, independent of model architecture or transfer learning strategy. Surprisingly, a simple baseline model that predicted site mean fitness matched or outperformed supervised models on many datasets, highlighting the dominant role of site effects. Analysis of site variability using two metrics, relative variability of site means (RVSM) and fraction of highly variable sites (FHVS), revealed that patterns of fitness variation within and among sites constrain model performance and largely explain the observed differences between viral and cellular datasets. Moreover, splitting training and test data by site, rather than pooling, revealed that supervised models often rely on site effects rather than capturing broader mutational patterns. These findings highlight limitations of current pLMs for mutational effect prediction and suggest that dataset composition, rather than model architecture or training, is the primary driver of predictive success. Significance StatementMutational effects prediction with protein language models tends to vary widely in prediction accuracy, depending on the dataset considered. While poor performance is commonly equated with poor model quality, we show here that intrinsic dataset features, such as the variability of fitness values within and among sites, are critical predictors of model performance. Moreover, we show that many existing benchmarks overestimate model performance, by allowing training data to leak into the test set. In fact, in many cases, protein language models barely outperform a naive predictor relying entirely on mean fitness values at individual sites. In aggregate, our study reveals that protein language models are not as reliable for predicting mutational effects as is commonly thought.

bioinformatics↗

bioRxiv2026

PeptideMTR: Scaling SMILES-Based Language Models for Therapeutic Peptide Engineering

Therapeutic peptides occupy a unique middle ground in drug discovery, offering the high specificity of protein interactions with the chemical diversity of small molecules, yet they currently fall in a computational blind spot. Existing foundation models cannot handle them effectively: protein models are restricted to natural amino acids, while chemical models struggle to process large, polymer-like sequences. This disconnect has forced the field to rely on static chemical descriptors that fail to capture subtle chemical details or on complex multi-embedding pipelines that are custom tailored to specific datasets. To bridge this gap, we present PeptideCLM-2, a suite of chemical language models trained on over 100 million molecules to natively represent complex peptide chemistry. This modeling approach expands the available toolkit of machine learning models for therapeutic peptides. Benchmarking results show strong performance versus prior methods for predicting development endpoints including membrane diffusion, biological function, and half life.

bioinformatics↗

bioRxiv2025

A fold switch regulates conformation of an alphavirus virus RNA-dependent RNA polymerase

Alphaviruses are mosquito-vectored, positive-strand RNA viruses causing rheumatic and neurological diseases. Like all RNA viruses, they encode an RNA-dependent RNA polymerase (RdRp, nsP4). Purification of an nsP4 derivative capable of processive RNA synthesis from a heteropolymeric template has been unsuccessful. Prior studies indicated Onyong-nyong virus (ONNV) nsP4 is soluble and requires additional non-structural proteins for activity. We performed biochemical and biophysical characterization of ONNV nsP4, including analytical ultracentrifugation and small-angle X-ray scattering (SAXS), revealing an extended conformation inconsistent with AlphaFold predictions of a compact structure. Fold switching was required for the extended conformation. Hydrogen-deuterium exchange mass spectrometry confirmed the fold-switched, extended state. Phylogenetic analysis showed conservation of residues contributing to both extended and compact states, implying functional roles for each. The extended form exhibited weak RNA binding and no polymerase activity on primed templates. The SAXS envelope of a precursor containing 50 amino acids from the nsP3 C-terminus (CT50-P34) matched the compact state. We propose precursor forms adopt the compact conformation. At the replication site, proteolytic cleavage would convert the precursor to an active polymerase. Polymerase dissociation upon completion of synthesis would induce fold switching to the inactive, extended state, precluding cytoplasmic activity that would activate intracellular immune responses.

biophysics↗

bioRxiv2025

Availability of charged tRNAs drives maximal protein synthesis at intermediate levels of codon usage bias

Synonymous codon usage can influence protein expression, since codons with high numbers of corresponding tRNAs are naturally translated more rapidly than codons with fewer corresponding tRNAs. Although translation efficiency ultimately depends on the concentration of aminoacylated (charged) tRNAs, many theoretical models of translation have ignored tRNA dynamics and treated charged tRNAs as fixed resources. This simplification potentially limits these models from making accurate predictions in situations where charged tRNAs become limiting. Here, we derive a mathematical model of translation with explicit tRNA dynamics and tRNA re-charging, based on a stochastic simulation of this system that was previously applied to investigate codon usage in the context of gene overexpression. We use the mathematical model to systematically explore the relationship between codon usage and the protein expression rate, and find that in the regime where tRNA charging is a limiting reaction, it is always optimal to match codon frequencies to the tRNA pool. Conversely, when tRNA charging is not limiting, using 100% of the preferred codon is optimal for protein production. We also use the tRNA dynamics model to augment a wholecell simulation of bacteriophage T7. Using this model, we demonstrate that the high expression rate of the T7 major capsid gene causes rare charged tRNAs to become entirely depleted, which explains the sensitivity of the major capsid gene to codon deoptimization.

biochemistry↗

bioRxiv2024

A Linear Mixed Effects Model for Evaluating Synthetic Gene Circuits

A significant advancement in synthetic biology is the development of synthetic gene circuits with predictive Boolean logic. However, there is no universally accepted or applied statistical test to analyze the performance of these circuits. Many basic statistical tests fail to capture the predicted logic (OR, AND, etc.) and most studies neglect statistical analysis entirely. As synthetic gene circuits shift toward advanced applications, primarily in computing, biosensing, and human health, it is critical to standardize the statistical methods used to evaluate gate success. Here, we propose the application of a linear mixed effects model to analyze and quantify genetic Boolean logic gate performance. First, we analyzed 144 currently published Boolean logic gates for trends and used unsupervised machine learning (k-means clustering) to validate the statistical model. Next, we utilized the model to generate estimates for the fixed effect of the ON state, {beta}, as a general descriptor of the Boolean nature of a circuit and used Monte Carlo simulations to recommend sample sizes for evaluating gate performance. Finally, we examined {beta} as a holistic metric for circuit performance using a series of nested repressor OR gates with intentionally degraded performance. We observed a linear correlation between {beta} and the predicted translation rate, highlighting the use of {beta} for the forward design of new Boolean gates. In summary, we utilized a linear mixed effects model to describe synthetic gene circuits and determined that the fixed effect, {beta}, is an appropriate descriptor of gate behavior that can be used to statistically evaluate performance. Significance statementThere is no standard method for statistically evaluating the success of biological Boolean logic gates and common statistical tests are not used or incorrectly applied. We propose the use of a linear mixed effects model to estimate the value for {beta}, a parameter which describes the difference between ON and OFF gate states. A null hypothesis of {beta} = 0 can then be tested to determine if there is a statistically significant difference between the two states.

synthetic biology↗

bioRxiv2024

Modeling and measuring how codon usage modulates the relationship between burden and yield during protein overexpression in bacteria

Excess utilization of translational resources is a critical source of burden on cells engineered to over-express exogenous proteins. To improve protein yields and genetic stability, researchers often use codon optimization strategies that improve translational efficiency by matching an exogenous genes codon usage with that of the host organisms highly expressed genes. Despite empirical data that shows the benefits of codon optimization, little is known quantitatively about the relationship between codon usage bias and the burden imposed by protein overexpression. Here, we develop and experimentally evaluate a stochastic gene expression model that considers the impact of codon usage bias on the availability of ribosomes and different tRNAs in a cell. In agreement with other studies, our model shows that increasing exogenous protein expression decreases production of native cellular proteins in a linear fashion. We also find that the slope of this relationship is modulated by how well the codon usage bias of the exogenous gene and the hosts genes match. Strikingly, we predict that an overoptimization domain exists where further increasing usage of optimal codons worsens yield and burden. We test our model by expressing sfGFP and mCherry2 from constructs that have a wide range of codon optimization levels in Escherichia coli. The results agree with our model, including for an mCherry2 gene sequence that appears to lose expression and genetic stability from codon overoptimization. Our findings can be leveraged by researchers to predict and design more optimal cellular systems through the use of more nuanced codon optimization strategies.

synthetic biology↗

bioRxiv2024

Scaling Down for Efficiency: Medium-Sized Transformer Models for Protein Sequence Transfer Learning

Protein language models (pLMs) can offer deep insights into evolutionary and structural properties of proteins. While larger models, such as the 15 billion parameter model ESM-2, promise to capture more complex patterns in sequence space, they also present practical challenges due to their high dimensionality and high computational cost. We systematically evaluated the performance of various ESM-style models across multiple biological datasets to assess the impact of model size on transfer learning via feature extraction. Surprisingly, we found that larger models do not necessarily outperform smaller ones, in particular when data is limited. Medium-sized models, such as ESM-2 650M and ESM C 600M, demonstrated consistently good performance, falling only slightly behind their larger counterparts--ESM-2 15B and ESM C 6B--despite being many times smaller. Additionally, we compared various methods of compressing embeddings prior to transfer learning, and we found that mean embeddings consistently outperformed other compression methods. In summary, ESM C 600M with mean embeddings offers an optimal balance between performance and efficiency, making it a practical and scalable choice for transfer learning in realistic biological applications.

biochemistry↗

bioRxiv2024

Single-cell virology: On-chip, quantitative characterization of the dynamics of virus spread from one single cell to another

Virus spread at the single-cell level is largely uncharacterized. We have designed and constructed a microfluidic device in which each nanowell contained a single, infected cell (donor) and a single, uninfected cell (recipient). Using a GFP-expressing poliovirus as our model, we observed both lytic and non-lytic spread. Donor cells supporting lytic spread established infection earlier than those supporting non-lytic spread. However, non-lytic spread established infections in recipient cells substantially faster than lytic spread and yielded higher rates of genome replication. While lytic spread was sensitive to the presence of capsid entry/uncoating inhibitors, non-lytic spread was not. Consistent with emerging models for non-lytic spread of enteroviruses using autophagy, reduction of LC3 levels in cells impaired non-lytic spread and elevated the fraction of virus in donor cells spreading lytically. The ability to distinguish lytic and non-lytic spread unambiguously will enable discovery of viral and host factors and host pathways used for non-lytic spread of enteroviruses and other viruses as well.

cell biology↗

bioRxiv2024

A systematic evaluation of the language-of-viral-escape model using multiple machine learning frameworks

Predicting the evolutionary patterns of emerging and endemic viruses is key for mitigating their spread in host populations. In particular, it is critical to rapidly identify mutations with the potential for immune escape or increased disease burden (variants of concern). Knowing which circulating mutations are such variants of concern can inform treatment or mitigation strategies such as alternative vaccines or targeted social distancing. A recent study proposed that variants of concern can be identified using two quantities extracted from protein language models, grammaticality and semantic change. These quantities are defined in analogy to concepts from natural language processing. Grammaticality is intended to be a measure of whether a variant viral protein is viable, and semantic change is intended to be a measure of potential for immune escape. Here, we systematically test this hypothesis, taking advantage of several high-throughput datasets that have become available, and also testing additional machine learning models for calculating the grammaticality metric. We find that grammaticality can be a measure of protein viability, though the more traditional metric {Delta}{Delta}G appears to be more effective. By contrast, we do not find compelling evidence that semantic change is a useful tool for identifying immune escape mutations.

bioinformatics↗

bioRxiv2024

Peptide-specific chemical language model successfully predicts membrane diffusion of cyclic peptides

Language modeling applied to biological data has significantly advanced the prediction of membrane penetration for small molecule drugs and natural peptides. However, accurately predicting membrane diffusion for peptides with pharmacologically relevant modifications remains a substantial challenge. Here, we introduce PeptideCLM, a peptide-focused chemical language model capable of encoding peptides with chemical modifications, unnatural or non-canonical amino acids, and cyclizations. We assess this model by predicting membrane diffusion of cyclic peptides, demonstrating greater predictive power than existing chemical language models. Our model is versatile and can be extended beyond membrane diffusion predictions to other target values. Its advantages include the ability to model macromolecules using chemical string notation, a largely unexplored domain, and a simple, flexible architecture that allows for adaptation to any peptide or other macromolecule dataset.

bioinformatics↗

bioRxiv2024

A computational model for bacteriophage {varphi}X174 gene expression

Bacteriophage{phi} X174 has been widely used as a model organism to study fundamental processes in molecular biology. However, several aspects of{phi} X174 gene regulation are not fully resolved. Here we construct a computational model for{phi} X174 and use the model to study gene regulation during the phage infection cycle. We estimate the relative strengths of transcription regulatory elements (promoters and terminators) by fitting the model to transcriptomics data. We show that the specific arrangement of a promoter followed immediately by a terminator, which occurs naturally in the{phi} X174 genome, poses a parameter identifiability problem for the model, since the activity of one element can be partially compensated for by the other. We also simulate{phi} X174 gene expression with two additional, putative transcription regulatory elements that have been proposed in prior studies. We find that the activities of these putative elements are estimated to be weak, and that variation in{phi} X174 transcript abundances can be adequately explained without them. Overall, our work demonstrates that{phi} X174 gene regulation is well described by the canonical set of promoters and terminators widely used in the literature.

microbiology↗

bioRxiv2023

Semantic search using protein large language models detects class II microcins in bacterial genomes

Class II microcins are antimicrobial peptides that have shown some potential as novel antibiotics. However, to date only ten class II microcins have been described, and discovery of novel microcins has been hampered by their short length and high sequence divergence. Here, we ask if we can use numerical embeddings generated by protein large language models to detect microcins in bacterial genome assemblies and whether this method can outperform sequence-based methods such as BLAST. We find that embeddings detect known class II microcins much more reliably than does BLAST and that any two microcins tend to have a small distance in embedding space even though they typically are highly diverged at the sequence level. In datasets of Escherichia coli, Klebsiella spp., and Enterobacter spp. genomes, we further find novel putative microcins that were previously missed by sequence-based search methods.

microbiology↗

bioRxiv2023

Deep mutational scanning and machine learning uncover antimicrobial peptide features driving membrane selectivity

Antimicrobial peptides commonly act by disrupting bacterial membranes, but also frequently damage mammalian membranes. Deciphering the rules governing membrane selectivity is critical to understanding their function and enabling their therapeutic use. Past attempts to decipher these rules have failed because they cannot interrogate adequate peptide sequence variation. To overcome this problem, we develop deep mutational surface localized antimicrobial display (dmSLAY), which reveals comprehensive positional residue importance and flexibility across an antimicrobial peptide sequence. We apply dmSLAY to Protegrin-1, a potent yet toxic antimicrobial peptide, and identify thousands of sequence variants that positively or negatively influence its antibacterial activity. Further analysis reveals that avoiding large aromatic residues and eliminating disulfide bound cysteine pairs while maintaining membrane bound secondary structure greatly improves Protegrin-1 bacterial specificity. Moreover, dmSLAY datasets enable machine learning to expand our analysis to include over 5.7 million sequence variants and reveal full Protegrin-1 mutational profiles driving either bacterial or mammalian membrane specificity. Our results describe an innovative, high-throughput approach for elucidating antimicrobial peptide sequence-structure-function relationships which can inform synthetic peptide-based drug design.

biochemistry↗

bioRxiv2023

Sequence and structure based deep learning models represent different aspects of protein biochemistry

Deep learning models are seeing increased use as methods to predict mutational effects or allowed mutations in proteins. The models commonly used for these purposes include large language models (LLMs) and 3D Convolutional Neural Networks (CNNs). These two model types have very different architectures and are commonly trained on different representations of proteins. LLMs make use of the transformer architecture and are trained purely on protein sequences whereas 3D CNNs are trained on voxelized representations of local protein structure. While comparable overall prediction accuracies have been reported for both types of models, it is not known to what extent these models make comparable specific predictions and/or generalize protein biochemistry in similar ways. Here, we perform a systematic comparison of two LLMs and two structure-based models (CNNs) and show that the different model types have distinct strengths and weaknesses. The overall prediction accuracies are largely uncorrelated between the sequence- and structure-based models. Overall, the two structure-based models are better at predicting buried aliphatic and hydrophobic residues whereas the two LLMs are better at predicting solvent-exposed polar and charged amino acids. Finally, we find that a combined model that takes the individual model predictions as input can leverage these individual model strengths and results in significantly improved overall prediction accuracy.

biochemistry↗

bioRxiv2023

Growth-dependent gene expression variation influences the strength of codon usage biases

The most highly expressed genes in microbial genomes tend to use a limited set of synonymous codons, often referred to as "preferred codons." The existence of preferred codons is commonly attributed to selection pressures on various aspects of protein translation including accuracy and/or speed. However, gene expression is condition-dependent and even within single-celled organisms transcript and protein abundances can vary depending on a variety of environmental and other factors. Here, we show that growth rate-dependent expression variation is an important constraint that significantly influences the evolution of gene sequences. Using large-scale transcriptomic and proteomic data sets in Escherichia coli and Saccharomyces cerevisiae, we confirm that codon usage biases are strongly associated with gene expression but highlight that this relationship is most pronounced when gene expression measurements are taken during rapid growth conditions. Specifically, genes whose relative expression increases during periods of rapid growth have stronger codon usage biases than comparably expressed genes whose expression decreases during rapid growth conditions. These findings highlight that gene expression measured in any particular condition tells only part of the story regarding the forces shaping the evolution of microbial gene sequences. More generally, our results imply that microbial physiology during rapid growth is critical for explaining long-term translational constraints.

evolutionary biology↗

bioRxiv2023

Distinct horizontal transfer mechanisms for type I and type V CRISPR-associated transposons

CRISPR-associated transposons (CASTs) co-opt CRISPR-Cas proteins and Tn7-family transposons for RNA-guided vertical and horizontal transmission. CASTs encode minimal CRISPR arrays but cant acquire new spacers. Here, we show that CASTs instead co-opt defense-associated CRISPR arrays for horizontal transmission. A bioinformatic analysis shows that all CAST sub-types co-occur with defense-associated CRISPR-Cas systems. Using an E. coli quantitative transposition assay, we show that CASTs use CRISPR RNAs (crRNAs) from these defense systems for horizontal gene transfer. A high-resolution structure of the type I-F CAST-Cascade in complex with a type III-B crRNA reveals that Cas6 recognizes direct repeats via sequence-independent{pi} - {pi} interactions. In addition to using heterologous CRISPR arrays, type V CASTs can also transpose via a crRNA-independent unguided mechanism, even when the S15 co-factor is over-expressed. Over-expressing S15 and the trans-activating CRISPR RNA (tracrRNA) or a single guide RNA (sgRNA) reduces, but does not abrogate, off-target integration for type V CASTs. Exploiting new spacers in defense-associated CRISPR arrays explains how CASTs horizontally transfer to new hosts. More broadly, this work will guide further efforts to engineer the activity and specificity of CASTs for gene editing applications.

molecular biology↗