Biology subjects

Zador, A.

Publications and source records attributed to Zador, A..

11 recordsLinked to original sources

bioRxiv2025

TissueFormer: a neural network for labeling tissue from grouped single-cell RNA profiles

BackgroundSingle-cell RNA sequencing technologies have enabled unprecedented insights into gene expression and opened new pathways for diagnostics and tissue annotation. At present, most computational approaches for interpreting single-cell data predict labels or properties based on isolated single-cell transcriptomic profiles. This approach overlooks the cellular composition within a sample, which is often critical for inferring tissue identity or other sample-level phenotypes. ResultsTo address this limitation, we introduce TissueFormer, a Transformer-based neural network that infers population-level labels from groups of single-cell RNA profiles while retaining single-cell resolution. We applied TissueFormer to two tasks: predicting COVID-19 severity from single-cell RNA sequencing of blood samples, and predicting cortical area identity from spatial transcriptomic data in mouse brains. TissueFormer outperformed single-cell foundation models and machine learning methods applied to pseudobulk and cell type composition. ConclusionsTissueFormers higher performance promises more accurate diagnostics and enables the automated construction of high-resolution brain region maps in individual mice directly from spatial transcriptomic data. Applied to mice with developmental perturbations to visual input, these maps revealed a significant reduction in predicted visual cortex area, illustrating how individual differences in neuroanatomy can be quantified. More broadly, TissueFormer provides a framework for predicting any population-level phenotypes which are influenced by cellular diversity and tissue-level organization.

bioinformatics↗

bioRxiv2024

Stochastic Wiring of Cell Types Enhances Fitness by Generating Phenotypic Variability

The development of neural connectivity is a crucial biological process that gives rise to diverse brain circuits and behaviors. Neural development is a stochastic process, but this stochasticity is often treated as a nuisance to overcome rather than as a functional advantage. Here we use a computational model, in which connection probabilities between discrete cell types are genetically specified, to investigate the benefits of stochasticity in the development of neural wiring. We show that this model can be viewed as a generalization of a powerful class of artificial neural networks--Bayesian neural networks--where each network parameter is a sample from a distribution. Our results reveal that stochasticity confers a greater benefit in large networks and variable environments, which may explain its role in organisms with larger brains. Surprisingly, we find that the average fitness over a population of agents is higher than a single agent defined by the average connection probability. Our model reveals how developmental stochasticity, by inducing a form of non-heritable phenotypic variability, can increase the probability that at least some individuals will survive in rapidly changing, unpredictable environments. Our results suggest how stochasticity may be an important feature rather than a bug in neural development.

neuroscience↗

bioRxiv2022

Modular cell type organization of cortical areas revealed by in situ sequencing

The cortex is composed of neuronal types with diverse gene expression that are organized into specialized cortical areas. These areas, each with characteristic cytoarchitecture (Brodmann 1909; Vogt and Vogt 1919; Von Bonin 1947), connectivity (Zingg et al. 2014; Harris et al. 2019), and neuronal activity (Schwarz et al. 2008; Ferrarini et al. 2009; He et al. 2009; Meunier et al. 2010; Bertolero et al. 2015), are wired into modular networks (Zingg et al. 2014; Harris et al. 2019; Huang et al. 2020). However, it remains unclear whether cortical areas and their modular organization can be similarly defined by their transcriptomic signatures and how such signatures are established in development. Here we used BARseq, a high-throughput in situ sequencing technique, to interrogate the expression of 104 cell type marker genes in 10.3 million cells, including 4,194,658 cortical neurons over nine mouse forebrain hemispheres at cellular resolution. De novo clustering of gene expression in single neurons revealed transcriptomic types that were consistent with previous single-cell RNAseq studies(Yao et al. 2021a; Yao et al. 2021b). Gene expression and the distribution of fine-grained cell types vary along the contours of cortical areas, and the composition of transcriptomic types are highly predictive of cortical area identity. Moreover, areas with similar compositions of transcriptomic types, which we defined as cortical modules, overlap with areas that are highly connected, suggesting that the same modular organization is reflected in both transcriptomic signatures and connectivity. To explore how the transcriptomic profiles of cortical neurons depend on development, we compared the cell type distributions after neonatal binocular enucleation. Strikingly, binocular enucleation caused the cell type compositional profiles of visual areas to shift towards neighboring areas within the same cortical module, suggesting that peripheral inputs sharpen the distinct transcriptomic identities of areas within cortical modules. Enabled by the high-throughput, low-cost, and reproducibility of BARseq, our study provides a proof-of-principle for using large-scale in situ sequencing to reveal brain-wide molecular architecture and to understand its development.

neuroscience↗

bioRxiv2021

Encoding innate ability through a genomic bottleneck

AO_SCPLOWBSTRACTC_SCPLOWAnimals are born with extensive innate behavioral capabilities, which arise from neural circuits encoded in the genome. However, the information capacity of the genome is orders of magnitude smaller than that needed to specify the connectivity of an arbitrary brain circuit, indicating that the rules encoding circuit formation must fit through a "genomic bottleneck" as they pass from one generation to the next. Here we formulate the problem of innate behavioral capacity in the context of artificial neural networks in terms of lossy compression of the weight matrix. We find that several standard network architectures can be compressed by several orders of magnitude, yielding pre-training performance that can approach that of the fully-trained network. Interestingly, for complex but not for simple test problems, the genomic bottleneck algorithm also captures essential features of the circuit, leading to enhanced transfer learning to novel tasks and datasets. Our results suggest that compressing a neural circuit through the genomic bottleneck serves as a regularizer, enabling evolution to select simple circuits that can be readily adapted to important real-world tasks. The genomic bottleneck also suggests how innate priors can complement conventional approaches to learning in designing algorithms for artificial intelligence.

neuroscience↗

bioRxiv2020

Replicability of spatial gene expression atlas data from the adult mouse brain

BackgroundSpatial gene expression is particularly interesting in the mammalian brain, with the potential to serve as a link between many data types. However, as with any type of expression data, cross-dataset benchmarking of spatial data is a crucial first step. Here, we assess the replicability, with reference to canonical brain sub-divisions, between the Allen Institutes in situ hybridization data from the adult mouse brain (ABA) and a similar dataset collected using Spatial Transcriptomics (ST). With the advent of tractable spatial techniques, for the first time we are able to benchmark the Allen Institutes whole-brain, whole-transcriptome spatial expression dataset with a second independent dataset that similarly spans the whole brain and transcriptome. ResultsWe use LASSO, linear regression, and correlation-based feature selection in a supervised learning framework to classify expression samples relative to their assayed location. We show that Allen reference atlas labels are classifiable using transcription, but that performance is higher in the ABA than ST. Further, models trained in one dataset and tested in the opposite dataset do not reproduce classification performance bi-directionally. Finally, while an identifying expression profile can be found for a given brain area, it does not generalize to the opposite dataset. ConclusionsIn general, we found that canonical brain area labels are classifiable in gene expression space within dataset and that our observed performance is not merely reflecting physical distance in the brain. However, we also show that cross-platform classification is not robust. Emerging spatial datasets from the mouse brain will allow further characterization of cross-dataset replicability.

genomics↗

bioRxiv2020

A standardized and reproducible method to measure decision-making in mice

Progress in science requires standardized assays whose results can be readily shared, compared, and reproduced across laboratories. Reproducibility, however, has been a concern in neuroscience, particularly for measurements of mouse behavior. Here we show that a standardized task to probe decision-making in mice produces reproducible results across multiple laboratories. We designed a task for head-fixed mice that combines established assays of perceptual and value-based decision making, and we standardized training protocol and experimental hardware, software, and procedures. We trained 140 mice across seven laboratories in three countries, and we collected 5 million mouse choices into a publicly available database. Learning speed was variable across mice and laboratories, but once training was complete there were no significant differences in behavior across laboratories. Mice in different laboratories adopted similar reliance on visual stimuli, on past successes and failures, and on estimates of stimulus prior probability to guide their choices. These results reveal that a complex mouse behavior can be successfully reproduced across multiple laboratories. They establish a standard for reproducible rodent behavior, and provide an unprecedented dataset and open-access tools to study decision-making in mice. More generally, they indicate a path towards achieving reproducibility in neuroscience through collaborative open-science approaches.

neuroscience↗

bioRxiv2019

A Critique of Pure Learning: What Artificial Neural Networks can Learn from Animal Brains

AO_SCPCAPBSTRACTC_SCPCAPOver the last decade, artificial neural networks (ANNs), have undergone a revolution, catalyzed in large part by better tools for supervised learning. However, training such networks requires enormous data sets of labeled examples, whereas young animals (including humans) typically learn with few or no labeled examples. This stark contrast with biological learning has led many in the ANN community posit that instead of supervised paradigms, animals must rely instead primarily on unsupervised learning, leading the search for better unsupervised algorithms. Here we argue that much of an animals behavioral repertoire is not the result of clever learning algorithms--supervised or unsupervised--but arises instead from behavior programs already present at birth. These programs arise through evolution, are encoded in the genome, and emerge as a consequence of wiring up the brain. Specifically, animals are born with highly structured brain connectivity, which enables them learn very rapidly. Recognizing the importance of the highly structured connectivity suggests a path toward building ANNs capable of rapid learning.

neuroscience↗

bioRxiv2018

High-throughput mapping of mesoscale connectomes in individual mice

Comprehensive analysis of neuronal networks requires brain-wide measurement of connectivity, activity, and gene expression. Although high-throughput methods are available for mapping brain-wide activity and transcriptomes, comparable methods for mapping region-to-region connectivity remain slow and expensive because they require averaging across hundreds of brains. Here we describe BRICseq, which leverages DNA barcoding and sequencing to map connectivity from single individuals in a few weeks and at low cost. Applying BRICseq to the mouse neocortex, we find that region-to-region connectivity provides a simple bridge relating transcriptome to activity: The spatial expression patterns of a few genes predict region-to-region connectivity, and connectivity predicts activity correlations. We also exploited BRICseq to map the mutant BTBR mouse brain, which lacks a corpus callosum, and recapitulated its known connectopathies. BRICseq allows individual laboratories to compare how age, sex, environment, genetics and species affect neuronal wiring, and to integrate these with functional activity and gene expression.

neuroscience↗

bioRxiv2018

A computational framework for converting high-throughput DNA sequencing data into neural circuit connectivity.

There is growing interest in determining the connectivity of neural circuits-the "connectome"-at single neuron resolution. Most approaches to circuit mapping rely on either microscopy or physiology, but these approaches have very limited throughput. We have recently proposed BOINC (Barcoding of Individual Neuronal Connectivity), a radically different approach to connectivity mapping based on high-throughput DNA sequencing. Here we describe the set of computational algorithms that serve to convert sequencing data into neural connectivity. We apply our computational pipeline to the results of proof-of-principle experiments illustrating an implementation of BOINC based on pseudorabies virus (PRV). PRV is capable of traversing individual synapses and carry DNA barcodes from one cell to another. Using this high-throughput sequencing data, we obtain 456-by-486 connectivity matrix between putative neurons. An inexpensive high-throughput technique for establishing circuit connectivity at single neuron resolution would represent a major advance in neuroscience.

neuroscience↗

bioRxiv2017

Using high-throughput barcode sequencing to efficiently map connectomes

The function of a neural circuit is determined by the details of its synaptic connections. At present, the only available method for determining a neural wiring diagram with single synapse precision--a \"connectome\"--is based on imaging methods that are slow, labor-intensive and expensive. Here we present SYNseq, a method for converting the connectome into a form that can exploit the speed and low cost of modern high-throughput DNA sequencing. In SYNseq, each neuron is labeled with a unique random nucleotide sequence--an RNA \"barcode\"--which is targeted to the synapse using engineered proteins. Barcodes in pre- and postsynaptic neurons are then associated through protein-protein crosslinking across the synapse, extracted from the tissue, and then joined into a form suitable for sequencing. Although at present the inefficiency in our hands of barcode joining precludes the widespread application of this approach, we expect that with further development SYNseq will enable tracing of complex circuits at high speed and low cost.

neuroscience↗

bioRxiv2017

Network cloning using DNA barcodes

The connections between neurons determine the computations performed by both artificial and biological neural networks. Recently, we have proposed SYNseq, a method for converting the connectivity of a biological network into a form that can exploit the tremendous efficiencies of high-throughput DNA sequencing. In SYNseq, each neuron is tagged with a random sequence of DNA--a \"barcode\"--and synapses are represented as barcode pairs. SYNseq addresses the analysis problem, reducing a network into a suspension of barcode pairs. Here we formulate a novel and complementary synthesis problem: How can the suspension of barcode pairs be used to \"clone\" or copy the network back into an uninitialized tabula rasa network? Although this synthesis problem might be expected to be computationally intractable, we find that, surprisingly, this problem can be solved efficiently, using only neuron-local information. We present the \"one barcode one cell\" (OBOC) algorithm, which forces all barcodes of a given sequence to coalesce into the same neuron, and show that it converges in a number of steps that is a power law of the network size. Rapid and reliable network cloning with single synapse precision is thus theoretically possible.

neuroscience↗