Our duration estimation flexibly adapts to the statistical properties of the temporal context. Humans and non-human species exhibit a perceptual bias towards the mean of durations previously observed as well as serial dependence, a perceptual bias towards the duration of recently processed events. Here we asked whether those two phenomena arise from a unitary mechanism or reflect the operation of two distinct systems that adapt separately to the global and local statistics of the environment. We employed a set of duration reproduction tasks in which the target duration was sampled from distributions with different variances and means. The central tendency and serial dependence biases were jointly modulated by the range and the variance of the prior, and these effects were well-captured by a unitary mechanism model in which temporal expectancies are updated after each trial based on perceptual observations. Alternative models that assume separate mechanisms for global and local contextual effects failed to capture the empirical results.

Single cell ATAC-seq (scATAC-seq) enables the mapping of regulatory elements in fine-grained cell types. Despite this advance, analysis of the resulting data is challenging, and large scale scATAC-seq data are difficult to obtain and expensive to generate. This motivates a method to leverage information from previously generated large scale scATAC-seq or scRNA-seq data to guide our analysis of new scATAC-seq datasets. We analyze scATAC-seq data using latent Dirichlet allocation (LDA), a Bayesian algorithm that was developed to model text corpora, summarizing documents as mixtures of topics defined based on the words that distinguish the documents. When applied to scATAC-seq, LDA treats cells as documents and their accessible sites as words, identifying "topics" based on the cell type-specific accessible sites in those cells. Previous work used uniform symmetric priors in LDA, but we hypothesized that nonuniform matrix priors generated from LDA models trained on existing data sets may enable improved detection of cell types in new data sets, especially if they have relatively few cells. In this work, we test this hypothesis in scATAC-seq data from whole C. elegans nematodes and SHARE-seq data from mouse skin cells. We show that nonsymmetric matrix priors for LDA improve our ability to capture cell type information from small scATAC-seq datasets.

Convolutional neural networks (CNNs) can provide powerful and flexible models of neural sensory processing. However, the utility of CNNs in studying the auditory system has been limited by their requirement for large datasets and the complex response properties of single auditory neurons. To address these limitations, we developed a population encoding model: a CNN that simultaneously predicts activity of several hundred neurons recorded during presentation of a large set of natural sounds. This approach defines a shared spectro-temporal space and pools statistical power across neurons. Population models of varying architecture performed consistently and substantially better than traditional linear-nonlinear models on data from primary and non-primary auditory cortex. Moreover, population models were highly generalizable. The output layer of a model pre-trained on one population of neurons could be fit to data from novel single units, achieving performance equivalent to that of neurons in the original fit data. This ability to generalize suggests that population encoding models capture a complete representational space across neurons in an auditory cortical field.

The clinical approvals of KRAS G12C inhibitors have been a revolutionary advance in precision oncology, but response rates are often modest. To improve patient selection, we developed an integrated model to predict KRAS dependency. By integrating molecular profiles of a large panel of cell lines from the DEMETER2 dataset, we built a binary classifier to predict a tumor's KRAS dependency. Monte Carlo cross validation via ElasticNet within the training set was used to compare model performance and to tune parameters α and λ. The final model was then applied to the validation set. We validated the model with genetic depletion assays and an external dataset of lung cancer cells treated with a G12C inhibitor. We then applied the model to several Cancer Genome Atlas (TCGA) datasets. The final "K20" model contains 20 features, including expression of 19 genes and KRAS mutation status. In the validation cohort, K20 had an AUC of 0.94 and accurately predicted KRAS dependency in both mutant and KRAS wild-type cell lines following genetic depletion. It was also highly predictive across an external dataset of lung cancer lines treated with KRAS G12C inhibition. When applied to TCGA datasets, specific subpopulations such as the invasive subtype in colorectal cancer and copy number high pancreatic adenocarcinoma were predicted to have higher KRAS dependency. The K20 model has simple yet robust predictive capabilities that may provide a useful tool to select patients with KRAS mutant tumors that are most likely to respond to direct KRAS inhibitors.

Gene editing characterization with currently available tools does not always give precise relative proportions among the different types of gene edits present in an edited bulk of cells. We have developed CRISPR-Analytics, CRISPR-A, which is a comprehensive and versatile genome editing web application tool and a nextflow pipeline to give support to gene editing experimental design and analysis. CRISPR-A provides a robust gene editing analysis pipeline composed of data analysis tools and simulation. It achieves higher accuracy than current tools and expands the functionality. The analysis includes mock-based noise correction, spike-in calibrated amplification bias reduction, and advanced interactive graphics. This expanded robustness makes this tool ideal for analyzing highly sensitive cases such as clinical samples or experiments with low editing efficiencies. It also provides an assessment of experimental design through the simulation of gene editing results. Therefore, CRISPR-A is ideal to support multiple kinds of experiments such as double-stranded DNA break-based engineering, base editing (BE), primer editing (PE), and homology-directed repair (HDR), without the need of specifying the used experimental approach.

In the plant sciences, results of laboratory studies often do not translate well to the field. To help close this lab-field gap, we developed a strategy for studying the wiring of plant traits directly in the field, based on molecular profiling and phenotyping of individual plants. Here, we use this single-plant omics strategy on winter-type Brassica napus (rapeseed). We investigate to what extent early and late phenotypes of field-grown rapeseed plants can be predicted from their autumnal leaf gene expression, and find that autumnal leaf gene expression not only has substantial predictive power for autumnal leaf phenotypes but also for final yield phenotypes in spring. Many of the top predictor genes are linked to developmental processes known to occur in autumn in winter-type B. napus accessions, such as the juvenile-to-adult and vegetative-to-reproductive phase transitions, indicating that the yield potential of winter-type B. napus is influenced by autumnal development. Our results show that single-plant omics can be used to identify genes and processes influencing crop yield in the field.

We present a machine learning-based interpretive framework (whatprot) for analyzing single molecule protein sequencing data produced by fluorosequencing, a recently developed proteomics technology that determines sparse amino acid sequences for many individual peptide molecules in a highly parallelized fashion. Whatprot uses Hidden Markov Models (HMMs) to represent the states of each peptide undergoing the various chemical processes during fluorosequencing, and applies these in a Bayesian classifier, in combination with pre-filtering by a k-Nearest Neighbors (kNN) classifier trained on large volumes of simulated fluorosequencing data. We have found that by combining the HMM based Bayesian classifier with the kNN pre-filter, we are able to retain the benefits of both, achieving both tractable runtimes and acceptable precision and recall for identifying peptides and their parent proteins from complex mixtures, outperforming the capabilities of either classifier on its own. Whatprot's hybrid kNN-HMM approach enables the efficient interpretation of fluorosequencing data using a full proteome reference database and should now also enable improved sequencing error rate estimates.

Persistent homology (PH) is a popular tool for topological data analysis that has found applications across diverse areas of research. It provides a rigorous method to compute robust topological features in discrete experimental observations that often contain various sources of uncertainties. Although powerful in theory, PH suffers from high computation cost that precludes its application to large data sets. Additionally, most analyses using PH are limited to computing the existence of nontrivial features. Precise localization of these features is not generally attempted because, by definition, localized representations are not unique and because of even higher computation cost. Such a precise location is a sine qua non for determining functional significance, especially in biological applications. Here, we provide a strategy and algorithms to compute tight representative boundaries around nontrivial robust features in large data sets. To showcase the efficiency of our algorithms and the precision of computed boundaries, we analyze the human genome and protein crystal structures. In the human genome, we found a surprising effect of the impairment of chromatin loop formation on loops through chromosome 13 and the sex chromosomes. We also found loops with long-range interactions between functionally related genes. In protein homologs with significantly different topology, we found voids attributable to ligand-interaction, mutation, and differences between species.

Luminal-A breast cancer is the most frequently occurring subtype which is characterized by high expression levels of hormone receptors. However, some luminal-A breast cancer patients suffer from intrinsic and/or acquired resistance to endocrine therapies which are considered as first-line treatments for luminal-A breast cancer. This heterogeneity within luminal-A breast cancer has required a more precise stratification method. Hence, our study aims to identify prognostic subgroups of luminal-A breast cancer. In this study, we discovered two prognostic subgroups of luminal-A breast cancer (BPS-LumA and WPS-LumA) using deep autoencoders and gene expressions. The deep autoencoders were trained using gene expression profiles of 679 luminal-A breast cancer samples in the METABRIC dataset. Then, latent features of each samples generated from the deep autoencoders were used for K-Means clustering to divide the samples into two subgroups, and Kaplan-Meier survival analysis was performed to compare prognosis (recurrence-free survival) between them. As a result, the prognosis between the two subgroups were significantly different (p-value = 5.82E-05; log-rank test). This prognostic difference between two subgroups was validated using gene expression profiles of 415 luminal-A breast cancer samples in the TCGA BRCA dataset (p-value = 0.004; log-rank test). Notably, the latent features were superior to the gene expression profiles and traditional dimensionality reduction method in terms of discovering the prognostic subgroups. Lastly, we discovered that ribosome-related biological functions could be potentially associated with the prognostic difference between them using differentially expressed genes and co-expression network analysis. Our stratification method can be contributed to understanding a complexity of luminal-A breast cancer and providing a personalized medicine.

Viruses evolve in infected host populations, and host population dynamics affect viral evolution. RNA viruses with a short duration of infection and a high peak viral load, such as SARS-CoV-2, are maintained in human populations. By contrast, RNA viruses characterized by a long infection duration and a low peak viral load (e.g., borna disease virus) can be maintained in nonhuman populations, and the process of the evolution of persistent viruses has rarely been explored. Here, using a multi-level modeling approach including both individual-level virus infection dynamics and population-scale transmission, we consider virus evolution based on the host environment, specifically, the effect of the contact history of infected hosts. We found that, with a highly dense contact history, viruses with a high virus production rate but low accuracy are likely to be optimal, resulting in a short infectious period with a high peak viral load. In contrast, with a low-density contact history, viral evolution is toward low virus production but high accuracy, resulting in long infection durations with low peak viral load. Our study sheds light on the origin of persistent viruses and why acute viral infections but not persistent virus infection tends to prevail in human society.

In human and other metazoans, the determinants of replication origin location and strength are still elusive. Origins are licensed in G1 phase and fired in S phase of the cell cycle, respectively. It is debated which of these two temporally separate steps determines origin efficiency. Experiments can independently profile mean replication timing (MRT) and replication fork directionality (RFD) genome-wide. Such profiles contain information on multiple origins' properties and on fork speed. Due to possible origin inactivation by passive replication, however, observed and intrinsic origin efficiencies can markedly differ. Thus, there is a need for methods to infer intrinsic from observed origin efficiency, which is context-dependent. Here, we show that MRT and RFD data are highly consistent with each other but contain information at different spatial scales. Using neural networks, we infer an origin licensing landscape that, when inserted in an appropriate simulation framework, jointly predicts MRT and RFD data with unprecedented precision and underlies the importance of dispersive origin firing. We furthermore uncover an analytical formula that predicts intrinsic from observed origin efficiency combined with MRT data. Comparison of inferred intrinsic origin efficiencies with experimental profiles of licensed origins (ORC, MCM) and actual initiation events (Bubble-seq, SNS-seq, OK-seq, ORM) show that intrinsic origin efficiency is not solely determined by licensing efficiency. Thus, human replication origin efficiency is set at both the origin licensing and firing steps.

Microbial communities have vital roles in systems essential to human health and agriculture, such as gut and soil microbiomes, and there is growing interest in engineering designer consortia for applications in biotechnology (e.g., personalized probiotics, bioproduction of high-value products, biosensing). The capacity to monitor and model metabolite exchange in dynamic microbial consortia can provide foundational information important to understand the community level behaviors that emerge, a requirement for building novel consortia. Where experimental approaches for monitoring metabolic exchange are technologically challenging, computational tools can enable greater access to the fate of both chemicals and microbes within a consortium. In this study, we developed an in-silico model of a synthetic microbial consortia of sucrose-secreting Synechococcus elongatus PCC 7942 and Escherichia coli W. Our model was built on the NUFEB framework for Individual-based Modeling (IbM) and optimized for biological accuracy using experimental data. We showed that the relative level of sucrose secretion regulates not only the steady-state support for heterotrophic biomass, but also the temporal dynamics of consortia growth. In order to determine the importance of spatial organization within the consortium, we fit a regression model to spatial data and used it to accurately predict colony fitness. We found that some of the critical parameters for fitness prediction were inter-colony distance, initial biomass, induction level, and distance from the center of the simulation volume. We anticipate that the synergy between experimental and computational approaches will improve our ability to design consortia with novel function.

PNCK, or CAMK1b, is an understudied kinase of the calcium-calmodulin dependent kinase family which recently has been identified as a marker of cancer progression and survival in several large-scale multi-omics studies. The biology of PNCK and its relation to oncogenesis has also begun to be elucidated, with data suggesting various roles in DNA damage response, cell cycle control, apoptosis and HIF-1-alpha related pathways. To further explore PNCK as a clinical target, potent small-molecule molecular probes must be developed. Currently, there are no targeted small molecule inhibitors in pre-clinical or clinical studies for the CAMK family. Additionally, there exists no experimentally derived crystal structure for PNCK. We herein report a three-pronged chemical probe discovery campaign which utilized homology modeling, machine learning, virtual screening and molecular dynamics to identify small molecules with low-micromolar potency against PNCK activity from commercially available compound libraries. We report the discovery of a hit-series for the first targeted effort towards discovering PNCK inhibitors that will serve as the starting point for future medicinal chemistry efforts for hit-to-lead optimization of potent chemical probes.

Machine learning tools have proven useful across biological disciplines, allowing researchers to draw conclusions from large datasets, and opening up new opportunities for interpreting complex and heterogeneous biological data. Alongside the rapid growth of machine learning, there have also been growing pains: some models that appear to perform well have later been revealed to rely on features of the data that are artifactual or biased; this feeds into the general criticism that machine learning models are designed to optimize model performance over the creation of new biological insights. A natural question arises: how do we develop machine learning models that are inherently interpretable or explainable? In this manuscript, we describe the SWIF(r) reliability score (SRS), a method building on the SWIF(r) generative framework that reflects the trustworthiness of the classification of a specific instance. The concept of the reliability score has the potential to generalize to other machine learning methods. We demonstrate the utility of the SRS when faced with common challenges in machine learning including: 1) an unknown class present in testing data that was not present in training data, 2) systemic mismatch between training and testing data, and 3) instances of testing data that have missing values for some attributes. We explore these applications of the SRS using a range of biological datasets, from agricultural data on seed morphology, to 22 quantitative traits in the UK Biobank, and population genetic simulations and 1000 Genomes Project data. With each of these examples, we demonstrate how the SRS can allow researchers to interrogate their data and training approach thoroughly, and to pair their domain-specific knowledge with powerful machine-learning frameworks. We also compare the SRS to related tools for outlier and novelty detection, and find that it has comparable performance, with the advantage of being able to operate when some data are missing. The SRS, and the broader discussion of interpretable scientific machine learning, will aid researchers in the biological machine learning space as they seek to harness the power of machine learning without sacrificing rigor and biological insight.

The entry of influenza virus into the host cell requires fusion of its lipid envelope with the host membrane. It is catalysed by viral hemagglutinin protein, whose fragments called fusion peptides become inserted into the target bilayer and initiate its merging with the viral membrane. Isolated fusion peptides are already capable of inducing lipid mixing between liposomes. Years of studies indicate that upon membrane binding they form bend helical structure whose degree of opening fluctuates between tightly closed hairpin and an extended boomerang. The actual way in which they initiate fusion remains elusive. In this work we employ atomistic simulations of wild type and fusion inactive W14A mutant of influenza fusion peptides confined between two closely apposed lipid bilayers. We characterise peptide induced membrane perturbation and determine the potential of mean force for the formation of the first fusion intermediate, an interbilayer lipid bridge called stalk. Our results demonstrate two routes through which the peptides can lower free energy barrier towards fusion. The first one assumes peptides capability to adopt transmembrane configuration which subsequently promotes the creation of a stalk-hole complex. The second involves surface bound peptide configuration and proceeds owing to its ability to stabilise stalk by fitting into the region of extreme negative membrane curvature resulting from its formation. In both cases, the active peptide conformation corresponds to tight helical hairpin, whereas extended boomerang geometry appears to be unable to provide favourable thermodynamic effect. The latter observation offers plausible explanation for long known inactivity of boomerang-stabilising W14A mutation.

Chromosomes are arranged in distinct territories within the nucleus of animal cells. Recent experiments have shown that these territories overlap at their edges, suggesting partial mixing during interphase. Experiments that knock-down of condensin II proteins during interphase indicate increased chromosome mixing, which demonstrates control of the mixing. In this study, we use a generic polymer simulation to quantify the dynamics of chromosome mixing over time. We introduce the chromosome mixing index, which quantifies the mixing of distinct chromosomes in the nucleus. We find that the chromosome mixing index in a small confinement volume (as a model of the nucleus), increases as a power-law of the time, with the scaling exponent varying non-monotonically with self-interaction and volume fraction. By comparing the chromosome mixing index with both monomer subdiffusion due to (non-topological) intermingling of chromosomes as well as even slower reptation, we show that for relatively large volume fractions, the scaling exponent of the chromosome mixing index is related to Rouse dynamics for relatively weak chromosome attractions and to reptation for strong attractions. In addition, we extend our model to more realistically account for the situation of the Drosophila chromosome by including the heterogeneity of the polymers and their lengths to account for microphase separation of euchromatin and heterochromatin and their interactions with the nuclear lamina. We find that the interaction with the lamina further impedes chromosome mixing.

Lung adenocarcinoma (LUAD) is a deadly tumor with dynamic evolutionary process. Although much endeavors have been made in identifying the temporal patterns of cancer progression, it remains challenging to infer and interpret the molecular alterations associated with cancer development and progression. To this end, we developed a computational approach to infer the progression trajectory based on cross-sectional transcriptomic data. Analysis of the LUAD data using our approach revealed a linear trajectory with three different branches for malignant progression, and the results showed consistency in three independent cohorts. We used the progression model to elucidate the potential molecular events in LUAD progression. Further analysis showed that overexpression of BUB1B, BUB1 and BUB3 promoted tumor cell proliferation and metastases by disturbing the spindle assembly checkpoint (SAC) in the mitosis. Aberrant mitotic spindle checkpoint signaling appeared to be one of the key factors promoting LUAD progression. We found the inferred cancer trajectory allows to identify LUAD susceptibility genetic variations using genome-wide association analysis. This result shows the opportunity for combining analysis of candidate genetic factors with disease progression. Furthermore, the trajectory showed clear evident mutation accumulation and clonal expansion along with the LUAD progression. Understanding how tumors evolve and identifying mutated genes will help guide cancer management. We investigated the clonal architectures and identified distinct clones and subclones in different LUAD branches. Validation of the model in multiple independent data sets and correlation analysis with clinical results demonstrate that our method is effective and unbiased.

To address ongoing academic achievement gap, there is a need for more school-university partnerships promoting early access to STEM education. During summer 2020, members of our institute initiated QBio-EDGE (Quantitative Biology-Empowering Diversity and Growth in Education), an outreach program for high schools in Los Angeles. In the hope of contributing to increasing diversity in academia, QBio-EDGE aims to make STEM education more accessible for students from historically excluded communities by exposing them to scientific research and diverse scientist role models. This program is led by early career researchers (ECRs), i.e., undergraduate, graduate, and postdoctoral researchers. In our first year, the outreach activities took place during virtual learning, presenting challenges and opportunities within the program development. Here, we provide a practical guide outlining our outreach efforts, key factors we considered in the program development, and hurdles we overcame. Specifically, we describe how we assembled our diverse team, how we established trusting partnerships with participating schools, and how we designed engaging student-centered, problem-based classroom modules on quantitative biology and computational methods applications to understand living systems. We also discuss the importance of increased institutional support. We hope that this may inspire researchers at all career stages to engage with local schools by participating in science outreach, specifically in quantitative and computational fields. We challenge institutions to actively strengthen these efforts.

Peripheral nerve stimulation is being investigated as a therapeutic tool in several clinical scenarios. However, the adopted devices have restricted ability to obtain desired outcomes with tolerable off-target effects. Recent promising solutions are not yet employed in clinical practice due to complex required surgeries, lack of long-term stability, and implant invasiveness. Here, we aimed to design a neural interface to address these issues, specifically dimensioned for pudendal and sacral nerves to potentially target sexual, bladder, or bowel dysfunctions. We designed the adaptable intrafascicular radial electrode (AIR) through realistic computational models. They account for detailed human anatomy, inhomogeneous anisotropic conductance, following the trajectories of axons along curving and branching fascicles, and detailed biophysics of axons. The model was validated against available experimental data. Thanks to computationally efficient geometry-based selectivity estimations we informed the electrode design, optimizing its dimensions to obtain the highest selectivity while maintaining low invasiveness. We then compared the AIR with state-of-the-art electrodes, namely InterStim leads, multipolar cuffs and transversal intrafascicular multichannel electrodes (TIME). AIR, comprising a flexible substrate, surface active sites, and radially inserted intrafascicular needles, is designed to be implanted in a few standard steps, potentially enabling fast implants. It holds potential for repeatable stimulation outcomes thanks to its radial structural symmetry. When compared in-silico, AIR consistently outperformed cuff electrodes and InterStim leads in terms of recruitment threshold and stimulation selectivity. AIR performed similarly or better than a TIME, with quantified less invasiveness. Finally, we showed how AIR can adapt to different nerve sizes and varying shapes while maintaining high selectivity. The AIR electrode shows the potential to fill a clinical need for an effective peripheral nerve interface. Its high predicted performance in all the identified requirements was enabled by a model-based approach, readily applicable for the optimization of electrode parameters in any peripheral nerve stimulation scenario.

Single-pulse electrical stimulation in the nervous system, often called cortico-cortical evoked potential (CCEP) measurement, is an important technique to understand how brain regions interact with one another. Voltages are measured from implanted electrodes in one brain area while stimulating another with brief current impulses separated by several seconds. Historically, researchers have tried to understand the significance of evoked voltage polyphasic deflections by visual inspection, but no general-purpose tool has emerged to understand their shapes or describe them mathematically. We describe and illustrate a new technique to parameterize brain stimulation data, where voltage response traces are projected into one another using a semi-normalized dot product. The length of timepoints from stimulation included in the dot product is varied to obtain a temporal profile of structural significance, and the peak of the profile uniquely identifies the duration of the response. Using linear kernel PCA, a canonical response shape is obtained over this duration, and then single-trial traces are parameterized as a projection of this canonical shape with a residual term. Such parameterization allows for dissimilar trace shapes from different brain areas to be directly compared by quantifying cross-projection magnitudes, response duration, canonical shape projection amplitudes, signal-to-noise ratios, explained variance, and statistical significance. Artifactual trials are automatically identified by outliers in sub-distributions of cross-projection magnitude, and rejected. This technique, which we call "Canonical Response Parameterization" (CRP) dramatically simplifies the study of CCEP shapes, and may also be applied in a wide range of other settings involving event-triggered data.

Current agricultural practices facilitate emergence and spread of plant diseases through the wide use of monocultures. Host mixtures are a promising alternative for sustainable plant disease control. Their effectiveness can be partly explained by priming-induced cross-protection among plants. Priming occurs when plants are challenged with non-infective pathogen genotypes, resulting in increased resistance to subsequent infections by infective pathogen genotypes. We developed an epidemiological model to explore how mixing two distinct resistant varieties can reduce disease prevalence. We considered a pathogen population composed of three genotypes infecting either one or both varieties. We found that host mixtures should not contain an equal proportion of resistant plants, but a biased ratio (e.g. 80 : 20) to minimize disease prevalence. Counter-intuitively, the optimal ratio of resistant varieties should contain a lower proportion of the costliest resistance for the pathogen to break. This benefit is amplified by priming. This strategy also prevents the invasion of pathogens breaking all resistances.

Random walks on networks are widely used to model stochastic processes such as search strategies, transportation problems or disease propagation. A prominent example of such process is the dynamics of naive T cells within the lymph node while they are scanning for antigens. The observed T cells trajectories in small sub-volumes of the lymph node are well modeled as a random walk and they have been shown to follow the lymphatic conduit network as substrate for migration. One can then ask how does the connectivity patterns of the lymph node conduit network affect the T cells collective exploration behavior. In particular, does the network display properties that are uniform across the whole volume of the lymph node or can we distinguish some heterogeneities? We propose a workflow to accurately and efficiently define and compute these quantities on large networks, which enables us to characterize heterogeneities within a very large published dataset of Lymph Node Conduit Network. To establish the significance of our results, we compared the results obtained on the lymph node to null models of varying complexity. We identified significantly heterogeneous regions characterized as "remote regions" at the poles and next to the medulla, while a large portion of the network promotes uniform exploration by T cells.

Natural products are chemical compounds that form the basis of many therapeutics used in the pharmaceutical industry. In microbes, natural products are synthesized by groups of colocalized genes called biosynthetic gene clusters (BGCs). With advances in high-throughput sequencing, there has been an increase of complete microbial isolate genomes and metagenomes, from which a vast number of BGCs are undiscovered. Here, we introduce a self-supervised learning approach designed to identify and characterize BGCs from such data. To do this, we represent BGCs as chains of functional protein domains and train a masked language model on these domains. We assess the ability of our approach to detect BGCs and characterize BGC properties in bacterial genomes. We also demonstrate that our model can learn meaningful representations of BGCs and their constituent domains, detect BGCs in microbial genomes, and predict BGC product classes. These results highlight self-supervised neural networks as a promising framework for improving BGC prediction and classification.

Living cells utilize signaling pathways to sense, transduce, and process information. As the extracellular stimulation often has rich temporal characteristics which may govern dynamic cellular responses, it is important to quantify the rate of information flow through the signaling pathways. In this study, we used an epithelial cell line expressing a light-activatable FGF receptor and an ERK activity reporter to assess the ability of the MAPK/ERK pathway to transduce signal encoded in a sequence of pulses. By stimulating the cells with random light pulse trains, we demonstrated that the MAPK/ERK channel capacity is at least 6 bits per hour. The input reconstruction algorithm detects the light pulses with 1-min accuracy 5 min after their occurrence. The high information transmission rate may enable the pathway to coordinate multiple processes including cell movement and respond to rapidly varying stimuli such as chemoattracting gradients created by other cells.

Computationally designed multi-subunit assemblies have shown considerable promise for a variety of applications, including a new generation of potent vaccines. One of the major routes to such materials is rigid body sequence-independent docking of cyclic oligomers into architectures with point group or lattice symmetries. Current methods for docking and designing such assemblies are tailored to specific classes of symmetry and are difficult to modify for novel applications. Here we describe RPXDock, a fast, flexible, and modular software package for sequence-independent rigid-body protein docking across a wide range of symmetric architectures that is easily customizable for further development. RPXDock uses an efficient hierarchical search and a residue-pair transform (RPX) scoring method to rapidly search through multidimensional docking space. We describe the structure of the software, provide practical guidelines for its use, and describe the available functionalities including a variety of score functions and filtering tools that can be used to guide and refine docking results towards desired configurations.

[This corrects the article DOI: 10.1371/journal.pcbi.1010701.].

Variability is an intrinsic property of biological systems and is often at the heart of their complex behaviour. Examples range from cell-to-cell variability in cell signalling pathways to variability in the response to treatment across patients. A popular approach to model and understand this variability is nonlinear mixed effects (NLME) modelling. However, estimating the parameters of NLME models from measurements quickly becomes computationally expensive as the number of measured individuals grows, making NLME inference intractable for datasets with thousands of measured individuals. This shortcoming is particularly limiting for snapshot datasets, common e.g. in cell biology, where high-throughput measurement techniques provide large numbers of single cell measurements. We introduce a novel approach for the estimation of NLME model parameters from snapshot measurements, which we call filter inference. Filter inference uses measurements of simulated individuals to define an approximate likelihood for the model parameters, avoiding the computational limitations of traditional NLME inference approaches and making efficient inferences from snapshot measurements possible. Filter inference also scales well with the number of model parameters, using state-of-the-art gradient-based MCMC algorithms such as the No-U-Turn Sampler (NUTS). We demonstrate the properties of filter inference using examples from early cancer growth modelling and from epidermal growth factor signalling pathway modelling.

Animals rely on different decision strategies when faced with ambiguous or uncertain cues. Depending on the context, decisions may be biased towards events that were most frequently experienced in the past, or be more explorative. A particular type of decision making central to cognition is sequential memory recall in response to ambiguous cues. A previously developed spiking neuronal network implementation of sequence prediction and recall learns complex, high-order sequences in an unsupervised manner by local, biologically inspired plasticity rules. In response to an ambiguous cue, the model deterministically recalls the sequence shown most frequently during training. Here, we present an extension of the model enabling a range of different decision strategies. In this model, explorative behavior is generated by supplying neurons with noise. As the model relies on population encoding, uncorrelated noise averages out, and the recall dynamics remain effectively deterministic. In the presence of locally correlated noise, the averaging effect is avoided without impairing the model performance, and without the need for large noise amplitudes. We investigate two forms of correlated noise occurring in nature: shared synaptic background inputs, and random locking of the stimulus to spatiotemporal oscillations in the network activity. Depending on the noise characteristics, the network adopts various recall strategies. This study thereby provides potential mechanisms explaining how the statistics of learned sequences affect decision making, and how decision strategies can be adjusted after learning.

Coronaviruses (CoVs) use -1 programmed ribosomal frameshifting stimulated by RNA pseudoknots in the viral genome to control expression of enzymes essential for replication, making CoV pseudoknots a promising target for anti-coronaviral drugs. Bats represent one of the largest reservoirs of CoVs and are the ultimate source of most CoVs infecting humans, including those causing SARS, MERS, and COVID-19. However, the structures of bat-CoV frameshift-stimulatory pseudoknots remain largely unexplored. Here we use a combination of blind structure prediction followed by all-atom molecular dynamics simulations to model the structures of eight pseudoknots that, together with the SARS-CoV-2 pseudoknot, are representative of the range of pseudoknot sequences in bat CoVs. We find that they all share some key qualitative features with the pseudoknot from SARS-CoV-2, notably the presence of conformers with two distinct fold topologies differing in whether or not the 5' end of the RNA is threaded through a junction, and similar conformations for stem 1. However, they differed in the number of helices present, with half sharing the 3-helix architecture of the SARS-CoV-2 pseudoknot but two containing 4 helices and two others only 2. These structure models should be helpful for future work studying bat-CoV pseudoknots as potential therapeutic targets.

Enhancers are stretches of regulatory DNA that bind transcription factors (TFs) and regulate the expression of a target gene. Shadow enhancers are two or more enhancers that regulate the same target gene in space and time and are associated with most animal developmental genes. These multi-enhancer systems can drive more consistent transcription than single enhancer systems. Nevertheless, it remains unclear why shadow enhancer TF binding sites are distributed across multiple enhancers rather than within a single large enhancer. Here, we use a computational approach to study systems with varying numbers of TF binding sites and enhancers. We employ chemical reaction networks with stochastic dynamics to determine the trends in transcriptional noise and fidelity, two key performance objectives of enhancers. This reveals that while additive shadow enhancers do not differ in noise and fidelity from their single enhancer counterparts, sub- and superadditive shadow enhancers have noise and fidelity trade-offs not available to single enhancers. We also use our computational approach to compare the duplication and splitting of a single enhancer as mechanisms for the generation of shadow enhancers and find that the duplication of enhancers can decrease noise and increase fidelity, although at the metabolic cost of increased RNA production. A saturation mechanism for enhancer interactions similarly improves on both of these metrics. Taken together, this work highlights that shadow enhancer systems may exist for several reasons: genetic drift or the tuning of key functions of enhancers, including transcription fidelity, noise and output.

In humans, glycogen storage diseases result from metabolic inborn errors, and can lead to severe phenotypes and lethal conditions. Besides these rare diseases, glycogen is also associated to widely spread societal burdens such as diabetes. Glycogen is a branched glucose polymer synthesised and degraded by a complex set of enzymes. Over the past 50 years, the structure of glycogen has been intensively investigated. Yet, the interplay between the detailed three-dimensional glycogen structure and the related enzyme activity is only partially characterised and still to be fully understood. In this article, we develop a stochastic coarse-grained and spatially resolved model of branched polymer biosynthesis following a Gillespie algorithm. Our study largely focusses on the role of the branching enzyme, and first investigates the properties of the model with generic parameter values, before comparing it to in vivo experimental data in mice. It arises that the ratio of glycogen synthase over branching enzyme reaction rates drastically impacts the structure of the granule. We deeply investigate the mechanism of branching and parametrise it using distinct lengths. Not only do we consider various possible sets of values for these lengths, but also distinct rules to apply them. We show how combining various values for these lengths finely tunes glycogen macromolecular structure. Comparing the model with experimental data confirms that we can accurately reproduce glycogen chain length distributions in wild type mice. Additional granule properties obtained for this fit are also in good agreement with typically reported values in the experimental literature. Nonetheless, we find that the mechanism of branching must be more flexible than usually reported. Overall, our model provides a theoretical basis to quantify the effect that single enzymatic parameters, in particular of the branching enzyme, have on the chain length distribution. Our generic model and methods can be applied to any glycogen data set, and could in particular contribute to characterise the mechanisms responsible for glycogen storage disorders.

Inference of biological network structures is often performed on high-dimensional data, yet is hindered by the limited sample size of high throughput "omics" data typically available. To overcome this challenge, often referred to as the "small n, large p problem," we exploit known organizing principles of biological networks that are sparse, modular, and likely share a large portion of their underlying architecture. We present SHINE-Structure Learning for Hierarchical Networks-a framework for defining data-driven structural constraints and incorporating a shared learning paradigm for efficiently learning multiple Markov networks from high-dimensional data at large p/n ratios not previously feasible. We evaluated SHINE on Pan-Cancer data comprising 23 tumor types, and found that learned tumor-specific networks exhibit expected graph properties of real biological networks, recapture previously validated interactions, and recapitulate findings in literature. Application of SHINE to the analysis of subtype-specific breast cancer networks identified key genes and biological processes for tumor maintenance and survival as well as potential therapeutic targets for modulating known breast cancer disease genes.

The druggability of small-molecule binding sites can be significantly affected by protein motions and conformational changes. Ligand binding, protein dynamics and protein function have been shown to be closely interconnected in myosins. The breakthrough discovery of omecamtiv mecarbil (OM) has led to an increased interest in small molecules that can target myosin and modulate its function for therapeutic purposes (myosin modulators). In this work, we use a combination of computational methods, including steered molecular dynamics, umbrella sampling and binding pocket tracking tools, to follow the evolution of the OM binding site during the recovery stroke transition of human β-cardiac myosin. We found that steering two internal coordinates of the motor domain can recapture the main features of the transition and in particular the rearrangements of the binding site, which shows significant changes in size, shape and composition. Possible intermediate conformations were also identified, in remarkable agreement with experimental findings. The differences in the binding site properties observed along the transition can be exploited for the future development of conformation-selective myosin modulators.

Modelling the transmission dynamics of an infectious disease is a complex task. Not only it is difficult to accurately model the inherent non-stationarity and heterogeneity of transmission, but it is nearly impossible to describe, mechanistically, changes in extrinsic environmental factors including public behaviour and seasonal fluctuations. An elegant approach to capturing environmental stochasticity is to model the force of infection as a stochastic process. However, inference in this context requires solving a computationally expensive "missing data" problem, using data-augmentation techniques. We propose to model the time-varying transmission-potential as an approximate diffusion process using a path-wise series expansion of Brownian motion. This approximation replaces the "missing data" imputation step with the inference of the expansion coefficients: a simpler and computationally cheaper task. We illustrate the merit of this approach through three examples: modelling influenza using a canonical SIR model, capturing seasonality using a SIRS model, and the modelling of COVID-19 pandemic using a multi-type SEIR model.

Liquid condensate droplets with distinct compositions of proteins and nucleic acids are widespread in biological cells. While it is known that such droplets, or compartments, can regulate irreversible protein aggregation, their effect on reversible self-assembly remains largely unexplored. In this article, we use kinetic theory and solution thermodynamics to investigate the effect of liquid-liquid phase separation on the reversible self-assembly of structures with well-defined sizes and architectures. We find that, when assembling subunits preferentially partition into liquid compartments, robustness against kinetic traps and maximum achievable assembly rates can be significantly increased. In particular, both the range of solution conditions leading to productive assembly and the corresponding assembly rates can increase by orders of magnitude. We analyze the rate equation predictions using simple scaling estimates to identify effects of liquid-liquid phase separation as a function of relevant control parameters. These results may elucidate self-assembly processes that underlie normal cellular functions or pathogenesis, and suggest strategies for designing efficient bottom-up assembly for nanomaterials applications.

In complex natural environments, sensory systems are constantly exposed to a large stream of inputs. Novel or rare stimuli, which are often associated with behaviorally important events, are typically processed differently than the steady sensory background, which has less relevance. Neural signatures of such differential processing, commonly referred to as novelty detection, have been identified on the level of EEG recordings as mismatch negativity (MMN) and on the level of single neurons as stimulus-specific adaptation (SSA). Here, we propose a multi-scale recurrent network with synaptic depression to explain how novelty detection can arise in the whisker-related part of the somatosensory thalamocortical loop. The "minimalistic" architecture and dynamics of the model presume that neurons in cortical layer 6 adapt, via synaptic depression, specifically to a frequently presented stimulus, resulting in reduced population activity in the corresponding cortical column when compared with the population activity evoked by a rare stimulus. This difference in population activity is then projected from the cortex to the thalamus and amplified through the interaction between neurons of the primary and reticular nuclei of the thalamus, resulting in rhythmic oscillations. These differentially activated thalamic oscillations are forwarded to cortical layer 4 as a late secondary response that is specific to rare stimuli that violate a particular stimulus pattern. Model results show a strong analogy between this late single neuron activity and EEG-based mismatch negativity in terms of their common sensitivity to presentation context and timescales of response latency, as observed experimentally. Our results indicate that adaptation in L6 can establish the thalamocortical dynamics that produce signatures of SSA and MMN and suggest a mechanistic model of novelty detection that could generalize to other sensory modalities.

Neural computations emerge from local recurrent neural circuits or computational units such as cortical columns that comprise hundreds to a few thousand neurons. Continuous progress in connectomics, electrophysiology, and calcium imaging require tractable spiking network models that can consistently incorporate new information about the network structure and reproduce the recorded neural activity features. However, for spiking networks, it is challenging to predict which connectivity configurations and neural properties can generate fundamental operational states and specific experimentally reported nonlinear cortical computations. Theoretical descriptions for the computational state of cortical spiking circuits are diverse, including the balanced state where excitatory and inhibitory inputs balance almost perfectly or the inhibition stabilized state (ISN) where the excitatory part of the circuit is unstable. It remains an open question whether these states can co-exist with experimentally reported nonlinear computations and whether they can be recovered in biologically realistic implementations of spiking networks. Here, we show how to identify spiking network connectivity patterns underlying diverse nonlinear computations such as XOR, bistability, inhibitory stabilization, supersaturation, and persistent activity. We establish a mapping between the stabilized supralinear network (SSN) and spiking activity which allows us to pinpoint the location in parameter space where these activity regimes occur. Notably, we find that biologically-sized spiking networks can have irregular asynchronous activity that does not require strong excitation-inhibition balance or large feedforward input and we show that the dynamic firing rate trajectories in spiking networks can be precisely targeted without error-driven training algorithms.

Non-ribosomal peptide synthetase (NRPS) is a diverse family of biosynthetic enzymes for the assembly of bioactive peptides. Despite advances in microbial sequencing, the lack of a consistent standard for annotating NRPS domains and modules has made data-driven discoveries challenging. To address this, we introduced a standardized architecture for NRPS, by using known conserved motifs to partition typical domains. This motif-and-intermotif standardization allowed for systematic evaluations of sequence properties from a large number of NRPS pathways, resulting in the most comprehensive cross-kingdom C domain subtype classifications to date, as well as the discovery and experimental validation of novel conserved motifs with functional significance. Furthermore, our coevolution analysis revealed important barriers associated with re-engineering NRPSs and uncovered the entanglement between phylogeny and substrate specificity in NRPS sequences. Our findings provide a comprehensive and statistically insightful analysis of NRPS sequences, opening avenues for future data-driven discoveries.

Representing past, present and future locations is key for spatial navigation. Indeed, within each cycle of the theta oscillation, the population of hippocampal place cells appears to represent trajectories starting behind the current position of the animal and sweeping ahead of it. In particular, we reported recently that the position represented by CA1 place cells at a given theta phase corresponds to the location where animals were or will be located at a fixed time interval into the past or future assuming the animal ran at its typical, not the current, speed through that part of the environment. This coding scheme leads to longer theta trajectories, larger place fields and shallower phase precession in areas where animals typically run faster. Here we present a mechanistic computational model that accounts for these experimental observations. The model consists of a continuous attractor network with short-term synaptic facilitation and depression that internally generates theta sequences that advance at a fixed pace. Spatial locations are then mapped onto the active units via modified Hebbian plasticity. As a result, neighboring units become associated with spatial locations further apart where animals run faster, reproducing our earlier experimental results. The model also accounts for the higher density of place fields generally observed where animals slow down, such as around rewards. Furthermore, our modeling results reveal that an artifact of the decoding analysis might be partly responsible for the observation that theta trajectories start behind the animal's current position. Overall, our results shed light on how the hippocampal code might arise from the interplay between behavior, sensory input and predefined network dynamics.

The interface between processing internal goals and salient events in the environment involves various top-down processes. Previous studies have identified multiple brain areas for salience processing, including the salience network (SN), dorsal attention network, and the locus coeruleus-norepinephrine (LC-NE) system. However, interactions among these systems in salience processing remain unclear. Here, we simultaneously recorded pupillometry, EEG, and fMRI during an auditory oddball paradigm. The analyses of EEG and fMRI data uncovered spatiotemporally organized target-associated neural correlates. By modeling the target-modulated effective connectivity, we found that the target-evoked pupillary response is associated with the network directional couplings from late to early subsystems in the trial, as well as the network switching initiated by the SN. These findings indicate that the SN might cooperate with the pupil-indexed LC-NE system in the reset and switching of cortical networks, and shed light on their implications in various cognitive processes and neurological diseases.

The outbreak of the severe acute respiratory syndrome coronavirus 2 started in Wuhan, China, towards the end of 2019 and spread worldwide. The rapid spread of the disease can be attributed to many factors including its high infectiousness and the high rate of human mobility around the world. Although travel/movement restrictions and other non-pharmaceutical interventions aimed at controlling the disease spread were put in place during the early stages of the pandemic, these interventions did not stop COVID-19 spread. To better understand the impact of human mobility on the spread of COVID-19 between regions, we propose a hybrid gravity-metapopulation model of COVID-19. Our modeling framework has the flexibility of determining mobility between regions based on the distances between the regions or using data from mobile devices. In addition, our model explicitly incorporates time-dependent human mobility into the disease transmission rate, and has the potential to incorporate other factors that affect disease transmission such as facemasks, physical distancing, contact rates, etc. An important feature of this modeling framework is its ability to independently assess the contribution of each factor to disease transmission. Using a Bayesian hierarchical modeling framework, we calibrate our model to the weekly reported cases of COVID-19 in thirteen local health areas in Metro Vancouver, British Columbia (BC), Canada, from July 2020 to January 2021. We consider two main scenarios in our model calibration: using a fixed distance matrix and time-dependent weekly mobility matrices. We found that the distance matrix provides a better fit to the data, whilst the mobility matrices have the ability to explain the variance in transmission between regions. This result shows that the mobility data provides more information in terms of disease transmission than the distances between the regions.

Stimulation to local areas remarkably affects brain activity patterns, which can be exploited to investigate neural bases of cognitive function and modify pathological brain statuses. There has been growing interest in exploring the fundamental action mechanisms of local stimulation. Nevertheless, how noise amplitude, an essential element in neural dynamics, influences stimulation-induced brain states remains unknown. Here, we systematically examine the effects of local stimulation by using a large-scale biophysical model under different combinations of noise amplitudes and stimulation sites. We demonstrate that noise amplitude nonlinearly and heterogeneously tunes the stimulation effects from both regional and network perspectives. Furthermore, by incorporating the role of the anatomical network, we show that the peak frequencies of unstimulated areas at different stimulation sites averaged across noise amplitudes are highly positively related to structural connectivity. Crucially, the association between the overall changes in functional connectivity as well as the alterations in the constraints imposed by structural connectivity with the structural degree of stimulation sites is nonmonotonically influenced by the noise amplitude, with the association increasing in specific noise amplitude ranges. Moreover, the impacts of local stimulation of cognitive systems depend on the complex interplay between the noise amplitude and average structural degree. Overall, this work provides theoretical insights into how noise amplitude and network structure jointly modulate brain dynamics during stimulation and introduces possibilities for better predicting and controlling stimulation outcomes.

To survive, insects must effectively navigate odor plumes to their source. In natural plumes, turbulent winds break up smooth odor regions into disconnected patches, so navigators encounter brief bursts of odor interrupted by bouts of clean air. The timing of these encounters plays a critical role in navigation, determining the direction, rate, and magnitude of insects' orientation and speed dynamics. Disambiguating the specific role of odor timing from other cues, such as spatial structure, is challenging due to natural correlations between plumes' temporal and spatial features. Here, we use optogenetics to isolate temporal features of odor signals, examining how the frequency and duration of odor encounters shape the navigational decisions of freely-walking Drosophila. We find that fly angular velocity depends on signal frequency and intermittency-the fraction of time signal can be detected-but not directly on durations. Rather than switching strategies when signal statistics change, flies smoothly transition between signal regimes, by combining an odor offset response with a frequency-dependent novelty-like response. In the latter, flies are more likely to turn in response to each odor hit only when the hits are sparse. Finally, the upwind bias of individual turns relies on a filtering scheme with two distinct timescales, allowing rapid and sustained responses in a variety of signal statistics. A quantitative model incorporating these ingredients recapitulates fly orientation dynamics across a wide range of environments and shows that temporal novelty detection, when combined with odor motion detection, enhances odor plume navigation.

Demixing signals in transcranial videos of neuronal calcium flux across the cerebral hemispheres is a key step before mapping features of cortical organization. Here we demonstrate that independent component analysis can optimally recover neural signal content in widefield recordings of neuronal cortical calcium dynamics captured at a minimum sampling rate of 1.5×106 pixels per one-hundred millisecond frame for seventeen minutes with a magnification ratio of 1:1. We show that a set of spatial and temporal metrics obtained from the components can be used to build a random forest classifier, which separates neural activity and artifact components automatically at human performance. Using this data, we establish functional segmentation of the mouse cortex to provide a map of ~115 domains per hemisphere, in which extracted time courses maximally represent the underlying signal in each recording. Domain maps revealed substantial regional motifs, with higher order cortical regions presenting large, eccentric domains compared with smaller, more circular ones in primary sensory areas. This workflow of data-driven video decomposition and machine classification of signal sources can greatly enhance high quality mapping of complex cerebral dynamics.

Cancer chemotherapy combines multiple drugs, but predicting the effects of drug combinations on cancer cell proliferation remains challenging, even for simple in vitro systems. We hypothesized that by combining knowledge of single drug dose responses and cell state transition network dynamics, we could predict how a population of cancer cells will respond to drug combinations. We tested this hypothesis here using three targeted inhibitors of different cell cycle states in two different cell lines in vitro. We formulated a Markov model to capture temporal cell state transitions between different cell cycle phases, with single drug data constraining how drug doses affect transition rates. This model was able to predict the landscape of all three different pairwise drug combinations across all dose ranges for both cell lines with no additional data. While further application to different cell lines, more drugs, additional cell state networks, and more complex co-culture or in vivo systems remain, this work demonstrates how currently available or attainable information could be sufficient for prediction of drug combination response for single cell lines in vitro.

The number of published metagenome assemblies is rapidly growing due to advances in sequencing technologies. However, sequencing errors, variable coverage, repetitive genomic regions, and other factors can produce misassemblies, which are challenging to detect for taxonomically novel genomic data. Assembly errors can affect all downstream analyses of the assemblies. Accuracy for the state of the art in reference-free misassembly prediction does not exceed an AUPRC of 0.57, and it is not clear how well these models generalize to real-world data. Here, we present the Residual neural network for Misassembled Contig identification (ResMiCo), a deep learning approach for reference-free identification of misassembled contigs. To develop ResMiCo, we first generated a training dataset of unprecedented size and complexity that can be used for further benchmarking and developments in the field. Through rigorous validation, we show that ResMiCo is substantially more accurate than the state of the art, and the model is robust to novel taxonomic diversity and varying assembly methods. ResMiCo estimated 7% misassembled contigs per metagenome across multiple real-world datasets. We demonstrate how ResMiCo can be used to optimize metagenome assembly hyperparameters to improve accuracy, instead of optimizing solely for contiguity. The accuracy, robustness, and ease-of-use of ResMiCo make the tool suitable for general quality control of metagenome assemblies and assembly methodology optimization.

Learning skilled behaviors requires intensive practice over days, months, or years. Behavioral hallmarks of practice include exploratory variation and long-term improvements, both of which can be impacted by circadian processes. During weeks of vocal practice, the juvenile male zebra finch transforms highly variable and simple song into a stable and precise copy of an adult tutor's complex song. Song variability and performance in juvenile finches also exhibit circadian structure that could influence this long-term learning process. In fact, one influential study reported juvenile song regresses towards immature performance overnight, while another suggested a more complex pattern of overnight change. However, neither of these studies thoroughly examined how circadian patterns of variability may structure the production of more or less mature songs. Here we relate the circadian dynamics of song maturation to circadian patterns of song variation, leveraging a combination of data-driven approaches. In particular we analyze juvenile singing in learned feature space that supports both data-driven measures of song maturity and generative developmental models of song production. These models reveal that circadian fluctuations in variability lead to especially regressive morning variants even without overall overnight regression, and highlight the utility of data-driven generative models for untangling these contributions.