Biology subjects

Levine, J.

Publications and source records attributed to Levine, J..

18 recordsLinked to original sources

bioRxiv2026

Ensembles of in silico structures enable T cell peptide-MHC binding prediction

Adaptive immunity relies on T-cell receptor (TCR) recognition of peptides presented by the major histocompatibility complex (pMHC). Accurate prediction of TCR:pMHC binding pairs from sequence data remains a longstanding challenge in computational immunology, limiting the development of precision immunotherapies like cancer vaccines and adoptive cell therapies. Here, we present enFoldX (ensemble of Folded compleXes), a structure-based approach leveraging biophysical characterization of AlphaFold3-generated ensembles to classify TCR:pMHC sequence pairs as cognate versus non-cognate. Unlike previous methods reliant on only sequence data or a single, static predicted structure, enFoldX extracts features from an entire generated ensemble with a custom focus on the biophysical binding interface. Our model distinguishes T cell reactivity between peptides differing by a single amino acid substitution, the resolution required for cancer neoantigens, and generalizes to unseen peptides, MHCs, and TCRs, a major objective for artificial intelligence (AI) in immunology. Our performance on these crucial tasks demonstrates that diverse, structural sampling of biophysical interactions over an ensemble is fundamental for accurate AI-driven binding predictions and offers lessons for efficient future data generation to improve models. Our findings therefore offer a scalable framework to accelerate therapeutic binder design, and we provide access to a publicly available code repository.

immunology↗

bioRxiv2026

Unifying coexistence theory for ecological communities and pathogen strain competition

Predicting the outcome of species or pathogen strain competition is a fundamental aim in both community ecology and infectious disease dynamics. Recent work revealed major challenges in predicting strain co-circulation from ecological coexistence theory due to overcompensatory competition among pathogens for susceptible resources, which can prevent the re-invasion of other competing strains. This resource overcompensation is ubiquitous across host-pathogen systems, but not apparent in simple Lotka-Volterra competition system, highlighting fundamental differences between pathogen strain and species competition. To address this gap, we begin by deriving classical models of pathogen strain and species competition from a resource-consumer model. This generalization illustrates that the relative time scale between resource and consumer dynamics limits the degree of resource overcompensation and therefore dictates the outcome of stochastic competition. Moreover, by introducing a mathematical framework for quantifying pairwise and higher-order terms from general competition systems, we show that a simple, ecological competition model can accurately predict the equilibrium dynamics of strain competition. A case study of rotavirus strain competition reveals that the ability to predict the outcome of strain competition from ecological theory depends on the underlying cross immunity structure. This work synthesizes coexistence theory across two fields by providing a unifying framework for predicting the outcome of complex ecological competition.

ecology↗

bioRxiv2025

Web engine for tumor pathology image retrievals on massive scales

Hematoxylin and Eosin staining (H&E) is widely used in clinical practice, but efficient and versatile image retrieval tools are lacking. We developed the H&E Retrieval Engine (HERE, https://hereapp.ccr.cancer.gov) to analyze patient cases based on image similarities to database records. Using H&E image regions as input, HERE searches 21.2 terabytes of whole-slide images from multiple tumor histopathology cohorts through a 12.1-gigabyte memory index, and returns top images containing regions similar to the query. HERE scans high-resolution images in the database using accurate artificial intelligence encoding and ultra-efficient hierarchical skip indexing. HERE demonstrated performance superior to existing image retrieval tools based on blinded pathologist scoring using benchmark queries that represent key image features of human tumors. By pairing spatial transcriptomics with H&E images, HERE also enables retrieving image features from gene transcriptomics input and identifies molecular pathways associated with tumor histologies.

bioinformatics↗

bioRxiv2025

Gene-embedding-based prediction and functional evaluation of perturbation expression responses with PRESAGE

Understanding the impact of genetic perturbations on cellular behavior is crucial for biological research, but comprehensive experimental mapping remains infeasible. We introduce PRESAGE (Perturbation Response EStimation with Aggregated Gene Embeddings), a simple, modular, and interpretable framework that predicts perturbation-induced expression changes by integrating diverse knowledge sources via gene embeddings. PRESAGE transforms gene embeddings through an attention-based model to predict perturbation expression outcomes. To assess model performance, we introduce a comprehensive evaluation suite with novel functional metrics that move beyond traditional regression tasks, including measures of accuracy in effect size prediction, in identifying perturbations with similar expression profiles (phenocopy), and in prediction of perturbations with the strongest impact on specific gene set scores. PRESAGE outperforms existing methods in both classical regression metrics and our novel functional evaluations. Through ablation studies, we demonstrate that knowledge source selection is more critical for predictive performance than architectural complexity, with cross-system Perturb-seq data providing particularly strong predictive power. We also find that performance saturates quickly with training set size, suggesting that experimental design strategies might benefit from collecting sparse perturbation data across multiple biological systems rather than exhaustive profiling of individual systems. Overall, PRESAGE establishes a robust framework for advancing perturbation response prediction and facilitating the design of targeted biological experiments, significantly improving our ability to predict cellular responses across diverse biological systems.

bioinformatics↗

bioRxiv2024

A unified theory for predicting pathogen mutual invasibility and co-circulation

A key aim in the dynamics and control of infectious diseases is predicting competitive outcomes of pathogen interactions. Observed pathogen community structure indicates both considerable coexistence of related variants and spectacular instances of replacement, notably in seasonal influenza and SARS-CoV-2. However, an overall comparative quantitative framework for invasion and coexistence remains elusive. Inspired by modern ecological coexistence theory, we address this gap by developing pathogen invasion theory (PIT) and test the resulting framework against empirical systems. PIT predicts near-universal mutual invasibility across major pathogen systems, including seasonal influenza strains and SARS-CoV-2 variants. Predicting co-circulation from mutual invasibility further depends on the extent of overcompensatory susceptible depletion dynamics. Our analyses highlight the central role of immuno-epidemiological factors in determining pathogen coexistence and community structure.

ecology↗

bioRxiv2024

Global spatial potential for implementing land-based climate mitigation

Land-based mitigation strategies (LBMS) are critical to reducing climate change and will require large areas for their implementation. Yet few studies have considered how and where LBMS compete for land or are mutually compatible across Earths surface. We derived high resolution estimates of the spatial distribution of 19 different LBMS. We estimated 8.9 billion ha suitable for LBMS across the Earth, including 5.6 Bha suitable for only one of the studied strategies and 3.3 Bha suitable for multiple LBMS. We identified significant conflicts between better carbon management of existing land cover types, and restoration-based strategies such as reforestation. At the same time, several agricultural management LBMS were compatible over large areas, including for example, enhanced weathering and improved plantation rotations. Our analysis presents local stakeholders, communities, and governments with the range of LBMS options, and the opportunity costs associated with scaling up any given LBMS to reduce global climate change.

ecology↗

bioRxiv2022

Two motor neuron synergies, invariant across ankle joint angles, activate the triceps surae during plantarflexion

Recent studies have suggested that the nervous system generates movements by controlling groups of motor neurons (synergies) that do not always align with muscle anatomy. In this study, we determined whether these synergies are robust across tasks with different mechanical constraints. We identified motor neuron synergies using principal component analysis (PCA) and cross-correlations between smoothed discharge rates of motor neurons. In Part 1, we used simulations to validate these methods. The results suggested that PCA can accurately identify the number of common inputs and their distribution across active motor neurons. Moreover, the results confirmed that cross-correlation can separate pairs of motor neurons that receive common inputs from those that do not receive common inputs. In Part 2, sixteen individuals performed plantarflexion at three ankle angles while we recorded electromyographic signals from the gastrocnemius lateralis (GL) and medialis (GM) and the soleus (SOL) with grids of surface electrodes. PCA revealed two motor neuron synergies. These motor neuron synergies were relatively stable with no significant differences in the distribution of motor neuron weights across ankle angles (p=0.62). When the cross-correlation was calculated for pairs of motor units tracked across ankle angles, we observed that only 13.0% of pairs of motor units from GL and GM exhibited significant correlations of their smoothed discharge rates across angles, confirming the low level of common inputs between these muscles. Overall, these results highlight the modularity of movement control at the motor neuron level, suggesting a sensible reduction of computational resources for movement control. Key points summaryO_LIThe central nervous system may generate movements by activating groups of motor neurons (synergies) with common inputs. C_LIO_LIWe show here that two main sources of common inputs drive the motor neurons innervating the triceps surae muscles during isometric ankle plantarflexions. C_LIO_LIWe report that the distribution of these common inputs is globally invariant despite changing the mechanical constraints of the tasks, i.e., the ankle angle. C_LIO_LIThese results suggest the functional relevance of the modular organization of the central nervous system to control movements. C_LI

neuroscience↗

bioRxiv2022

Toward a generalizable deep CNN for neural drive estimation across muscles and participants

High-density electromyography (HD-EMG) decomposition algorithms are used to identify individual motor unit spike trains, which collectively constitute the neural code of movements, to predict motor intent. This approach has advanced from offline to online decomposition, from isometric to dynamic contractions, leading to a wide range of neural-machine interface applications. However, current online methods need offline retraining when applied to the same muscle on a different day or to a different person, which limits their applications in a real-time neural-machine interface. We proposed a deep convolutional neural network (CNN) framework for neural drive estimation, which captures general spatiotemporal properties of motor unit action potentials to generalize its application without retraining to HD-EMG data recorded in separate sessions, muscles, and participants. We recorded HD-EMG signals from the vastus medialis and vastus lateralis muscles while participants performed isometric contractions during two sessions separated by approximately 20 months. We identified motor unit spike trains from HD-EMG signals using a blind source separation (BSS) method, and then used the cumulative spike train (CST) of these motor units and the HD-EMG signals to train and validate the deep CNN. On average, the correlation coefficients between CST from BSS and that from deep CNN were 0.977{+/-}0.007 for leave-one-out across-sessions-and-muscles validation and 0.985{+/-}0.005 for leave-one-out across-participants validation. When trained with more than four datasets, the performance of deep CNN saturated at 0.979{+/-}0.001 for cross validations across muscles, sessions, and participants. Therefore, we can conclude that the deep CNN is generalizable across the afore-mentioned conditions without retraining. We could potentially generate a robust deep CNN to estimate neural drive to muscles for neural-machine interfaces.

neuroscience↗

bioRxiv2022

Efficient in vivo neuronal genome editing in the mouse brain using nanocapsules containing CRISPR-Cas9 ribonucleoproteins

Genome editing of somatic cells via clustered regularly interspaced short palindromic repeats (CRISPR) offers promise for new therapeutics to treat a variety of genetic disorders, including neurological diseases. However, the dense and complex parenchyma of the brain and the post-mitotic state of neurons make efficient genome editing challenging. In vivo delivery systems for CRISPR-Cas proteins and single guide RNA (sgRNA) include both viral vectors and non-viral strategies, each presenting different advantages and disadvantages for clinical application. We developed non-viral and biodegradable PEGylated nanocapsules (NCs) that deliver preassembled Cas9-sgRNA ribonucleoproteins (RNPs). Here, we show that the RNP NCs led to robust genome editing in neurons following intracerebral injection into the mouse striatum. Genome editing was predominantly observed in medium spiny neurons (>80%), with occasional editing in cholinergic, calretinin, and parvalbumin interneurons. Glial activation was minimal and was localized along the needle tract. Our results demonstrate that the RNP NCs are capable of safe and efficient neuronal genome editing in vivo. SIGNIFICANCE STATEMENTModifying the DNA of cells in the brain could present opportunities for new treatments of neurological diseases. In this report, we describe a nanocapsule system designed to deliver the elements needed to modify the DNA of brain cells, also known as genome editing. These nanocapsules are created by chemically encapsulating the genome editing components, such that the nanocapsules are stable when prepared and biodegradable to release their payload upon entering cells. When injected into the mouse brain, our research shows that the nanocapsules lead to safe and efficient editing of DNA in neurons.

neuroscience↗

bioRxiv2022

Pharmacological activation of the mitochondrial stress protease OMA1 reveals a therapeutic liability in Diffuse Large B-Cell Lymphoma

Constitutive activation of the ATF4-mediated integrated stress response (ATF4-ISR) is common in cancer and buffers the metabolic challenges imposed by rapid proliferation. However, hyperactivation of the ISR can induce apoptosis. Here we demonstrate that novel pyrazolo-thiazole derivates activate the mitochondrial protease OMA1 which subsequently induces apoptosis in diffuse large B-cell lymphoma (DLBCL) cells. Apoptosis is dependent on the OMA1 mediated cleavage of DELE1 which leads to activation of HRI and induction of the ATF4 ISR. Screening in 406 cancer cell lines identified an inverse correlation between sensitivity to OMA1 activators and expression of the mitochondrial protein FAM210B. Ectopic overexpression of FAM210B specifically blocks OMA1 activation and apoptosis induction by pyrazolo-thiazole activators in DLBCL. OMA1 activators, including the preclinical candidate BTM-3566, selectively killed ABC, GCB, and double-hit DLBCL lines and induced complete tumor regression across a panel of DLBCL patient-derived xenografts. SignificanceHere we describe a novel class of small molecules that activate the mitochondrial protease OMA1 and induce therapeutic responses in DLBCL preclinical models in vitro and in vivo. OMA1 activation drives apoptosis through ATF4-ISR, an orthogonal mechanism to current therapies.

cancer biology↗

bioRxiv2020

Human-associated microbiota suppress invadingbacteria even under disruption by antibiotics

In light of their adverse impacts on resident microbial communities, it is widely predicted that broad-spectrum antibiotics can promote the spread of resistance by releasing resistant strains from competition with other strains and species. We investigated the invasion success of a resistant strain of Escherichia coli inoculated into human-associated communities in the presence and absence of the broad and narrow spectrum antibiotics rifampicin and polymyxin B, respectively. We found strong evidence of community-level suppression of the resistant strain in the absence of antibiotics and, despite large changes in community composition and abundance following rifampicin exposure, suppression of the invading resistant strain was maintained in both antibiotic treatments. Instead, the strength of competitive suppression was more strongly associated with the individual donor from which the community was sampled. This suggests microbiome composition strongly influences susceptibility to invasion by antibiotic-resistant strains, but at least some antibiotic-associated disruption can be tolerated before invasion susceptibility increases. A deeper understanding of this association will aid the development of ecologically-aware strategies for managing antibiotic resistance.

ecology↗

bioRxiv2020

Antibiotics in microbial communities: an ecological frame of resistance

There is growing awareness that our ability to tackle antibiotic resistance is limited by a lack of mechanistic understanding of the communities in which resistant microbes are embedded. The widespread coexistence of resistant and sensitive bacteria in microbial systems presents an especially frustrating paradox. Recent advances in ecological coexistence theory offer a powerful framework to probe the mechanisms regulating intra- and inter-specific coexistence, but the significance of this body of theory to the problem of antimicrobial resistance has been largely overlooked. In this Perspectives article, we draw on emerging ecological theory to illustrate how changes in both competitive ability and niche overlap are critical for understanding costs of resistance and the persistence of pathogens in microbial systems. We then show how trade-offs in resource acquisition strategies can have counter-intuitive consequences for the coexistence of resistant and susceptible genotypes in variable environments. These insights highlight numerous opportunities for innovative experimental and theoretical research into antibiotic resistance in the microbiome.

ecology↗

bioRxiv2020

The complexity of model-free and model-based learning strategies

Different learning strategies are thought to fall along a continuum that ranges from simple, inflexible, and fast "model-free" strategies, to more complex, flexible, and deliberative "model-based strategies". Here we show that, contrary to this proposal, strategies at both ends of this continuum can be equally flexible, effective, and time-intensive. We analyzed behavior of adult human subjects performing a canonical learning task used to distinguish between model-free and model-based strategies. Subjects using either strategy showed similarly high information complexity, a measure of strategic flexibility, and comparable accuracy and response times. This similarity was apparent despite the generally higher computational complexity of model-based algorithms and fundamental differences in how each strategy learned: model-free learning was driven primarily by observed past responses, whereas model-based learning was driven primarily by inferences about latent task features. Thus, model-free and model-based learning differ in the information they use to learn but can support comparably flexible behavior. Statement of RelevanceThe distinction between model-free and model-based learning is an influential framework that has been used extensively to understand individual- and task-dependent differences in learning by both healthy and clinical populations. A common interpretation of this distinction that model-based strategies are more complex and therefore more flexible than model-free strategies. However, this interpretation conflates computational complexity, which relates to processing resources and generally higher for model-based algorithms, with information complexity, which reflects flexibility but has rarely been measured. Here we use a metric of information complexity to demonstrate that, contrary to this interpretation, model-free and model-based strategies can be equally flexible, effective, and time-intensive and are better distinguished by the nature of the information from which they learn. Our results counter common interpretations of model-free versus model-based learning and demonstrate the general usefulness of information complexity for assessing different forms of strategic flexibility.

neuroscience↗

bioRxiv2018

Ecosystem tipping points in an evolving world

There is growing concern over tipping points arising in ecosystems due to the crossing of environmental thresholds. Tipping points lead to strong and possibly irreversible shifts between alternative ecosystem states incurring high societal costs. Traits are central to the feedbacks that maintain alternative ecosystem states, as they govern the responses of populations to environmental change that could stabilize or destabilize ecosystem states. However, we know little about how evolutionary changes in trait distributions over time affect the occurrence of tipping points, and even less about how big scale ecological shifts reciprocally interact with trait dynamics. We argue that interactions between ecological and evolutionary processes should be taken into account for understanding the balance of feedbacks governing tipping points in nature.

ecology↗

bioRxiv2018

Chemical signatures of honey bee group membership develop via a socially-modulated innate process

Large social insect colonies exhibit a remarkable ability for recognizing group members via colony-specific cuticular hydrocarbon (CHC) pheromonal signatures. Previous work suggested that in some ant species colony-specific signatures are generated through a \"gestalt\" mechanism via the passive transfer and homogenization of CHCs across all individual members of the colony. In contrast, we demonstrate that nestmate recognition cues of worker honey bees (Apis mellifera) mature in foragers via a sequence of stereotypic age-dependent quantitative and qualitative chemical transitions, which are driven by intrinsic biosynthetic pathways. Therefore, in contrast to predictions of the \"Gestalt\" model, nestmate recognition cues in honey bee colonies do not represent a passive \"average\" signature that is carried and recognized by all colony members. Instead, specific colony members develop the relevant cues via an innately-determined developmental program that can be modulated by colony-specific social environmental factors.

animal behavior and cognition↗

bioRxiv2018

Palantir characterizes cell fate continuities in human hematopoiesis

Recent studies using single cell RNA-seq (scRNA-seq) data derived from differentiating systems have raised fundamental questions regarding the discrete vs continuous nature of both differentiation and cell fate. Here we present Palantir, an algorithm that models trajectories of differentiating cells, which treats cell-fate as a probabilistic process, and leverages entropy to measure the changing nature of cell plasticity along the differentiation trajectory. Palantir generates a high resolution pseudotime ordering of cells, and assigns each cell state with its probability to differentiate into each terminal state. We apply Palantir to human bone marrow scRNA-seq data and detect key landmarks of hematopoietic differentiation. Palantirs resolution enables identification of key transcription factors driving lineage fate choices, as these TFs closely track when cells lose plasticity. We demonstrate that Palantir is generalizable to diverse tissue types and well-suited to resolve less studied differentiating systems.

systems biology↗

bioRxiv2018

Tissue-specific cis-regulatory divergence implicates a fatty acid elongase necessary for inhibiting interspecies mating in Drosophila

Pheromones known as cuticular hydrocarbons are a major component of reproductive isolation in Drosophila. Individuals from morphologically similar sister species produce different sets of hydrocarbons that allow potential mates to identify them as a suitable partner. In order to explore the molecular mechanisms underlying speciation, we performed RNA-seq in F1 hybrids to measure tissue-specific cis-regulatory divergence between the sister species D. simulans and D. sechellia. By focusing on cis-regulatory changes specific to female oenocytes, we rapidly identified a small number of candidate genes. We found that one of these, the fatty acid elongase eloF, broadly affects both the complement of hydrocarbons present on D. sechellia females and the propensity of D. simulans males to mate with those females. In addition, knockdown of eloF in the more distantly related D. melanogaster led to a similar shift in hydrocarbons as well as lower interspecific mate discrimination by D. simulans males. Thus, cis-regulatory changes in eloF appear to be a major driver in the sexual isolation of D. simulans from multiple other species. More generally, our RNA-seq approach proved to be far more efficient than QTL mapping in identifying candidate genes; the same framework can be used to pinpoint cis-regulatory drivers of divergence in a wide range of traits differing between any interfertile species.

systems biology↗

bioRxiv2018

Can Drosophila melanogaster tell who’s who?

Drosophila melanogaster are known to live in a social but cryptic world of touch and odours, but the extent to which they can perceive and integrate visual information is a hotly debated topic. Some researchers fixate on the limited resolution of D. melanogasters optics, others on their seemingly identical appearance; yet there is evidence of individual recognition and surprising visual learning in flies. Here, we apply machine learning and show that individual D. melanogaster are visually distinct. We also use the striking similarity of Drosophilas visual system to current convolutional neural networks to theoretically investigate D. melanogasters capacity for visual understanding. We find that, despite their limited optical resolution, D. melanogasters neuronal architecture has the capability to extract and encode a rich feature set that allows flies to re-identify individual conspecifics with surprising accuracy. These experiments provide a proof of principle that Drosophila inhabit in a much more complex visual world than previously appreciated.\n\nAuthor summaryIn this paper, we determine a proof of principle for inter-individual recognition in two parts; is there enough information contained in low resolution pictures for inter-fly discrimination, and if so does Drosophilas visual system have enough capacity to use it. We show that the information contained in a 29x29 pixel image (number of ommatidia in a fly eye) is sufficient to achieve 94% accuracy in fly re-identification. Further, we show that the fly eye has the theoretical capacity to identify another fly with about 75% accuracy. Although it is unlikely that flies use the exact algorithm we tested, our results show that, in principle, flies may be using visual perception in ways that are not usually appreciated.

animal behavior and cognition↗