A5064939285- RNA and protein synthesis mechanisms
- Protein Structure and Dynamics
- Statistical Methods in Clinical Trials
- DNA and Nucleic Acid Chemistry
- Gaussian Processes and Bayesian Inference
- Single-cell and spatial transcriptomics
- Statistical Methods and Inference
- RNA Research and Splicing
- Control Systems and Identification
- Gene Regulatory Network Analysis
- Stochastic processes and financial applications
- Cell Image Analysis Techniques
- Advanced Fluorescence Microscopy Techniques
- Fault Detection and Control Systems
- Glioma Diagnosis and Treatment
- Machine Learning and Algorithms
- Point processes and geometric inequalities
- RNA modifications and cancer
- Bayesian Modeling and Causal Inference
- Advanced Data Compression Techniques
- Image and Signal Denoising Methods
- Stochastic processes and statistical mechanics
- Domain Adaptation and Few-Shot Learning
- stochastic dynamics and bifurcation
- Optimal Experimental Design Methods
University of Michigan
2023-2025
Columbia University
2018-2024
Brain (Germany)
2021
Center for Theoretical Physics
2021
Brown University
2019
Allen Institute for Brain Science
2019
City University of Seattle
2018
In variational inference (VI), an approximation of the posterior distribution is selected from a family distributions through numerical optimization. With most common objective function, known as evidence lower bound (ELBO), only convergence to local optimum can be guaranteed. this work, we instead establish global particular VI method. This method, which may considered instance neural estimation (NPE), minimizes expectation inclusive (forward) KL divergence fit that parameterized by...
SUMMARY The corticospinal tract exerts its influence on movement through spinal neurons, which can be divided into types that exhibit distinct functions. However, it remains unknown whether these functional distinctions are reflected in the inputs different of neurons receive. Using rabies monosynaptic tracing from individual neuron cervical cord and 3D histological reconstruction mice, we discovered receive distinctly distributed across cortex, aligned with cell type function. This included...
SUMMARY Here, we report the spatial organization of RNA transcription and associated enhancer dynamics in human spinal cord at single-cell single-molecule resolution. We expand traditional multiomic measurements to reveal epigenetically poised bivalent active transcriptional states that define cell type specification. Simultaneous detection chromatin accessibility histone modifications nuclei reveals previously unobserved cell-type specific cryptic activity, which activation is uncoupled...
Modern spatial transcriptomics methods can target thousands of different types RNA transcripts in a single slice tissue. Many biological applications demand high density relative to the imaging resolution, leading partial mixing transcript rolonies many voxels; unfortunately, current analysis do not perform robustly this highly-mixed setting. Here we develop new approach, BARcode DEmixing through Non-negative Spatial Regression (BarDensr): start with generative model physical process that...
Abstract Modern spatial transcriptomics methods can target thousands of different types RNA transcripts in a single slice tissue. Many biological applications demand high density relative to the imaging resolution, leading partial mixing transcript rolonies many voxels; unfortunately, current analysis do not perform robustly this highly-mixed setting. Here we develop new approach, BARcode DEmixing through Non-negative Spatial Regression (BarDensr): start with generative model physical...
Cellular barcoding methods offer the exciting possibility of ‘infinite-pseudocolor’ anatomical reconstruction—i.e., assigning each neuron its own random unique barcoded ‘pseudocolor,’ and then using these pseudocolors to trace microanatomy neuron. Here we use simulations, based on densely-reconstructed electron microscopy microanatomy, with signal structure matched real data, quantify feasibility this procedure. We develop a new blind demixing approach recover barcodes that label neuron,...
Approximately three times per second, human visual perception is interrupted by a saccadic eye movement. In addition to taking the eyes new location, several lines of evidence suggest that saccades play multiple roles in perception. Indeed, it may be crucial processing informed about movements order analyze input distinctly and efficiently on each fixation preserve stable world across saccades. A variety studies has demonstrated activity brain areas modulated The hypothesis tested here these...
Abstract Glioblastoma (GBM) is a lethal disease characterized by inevitable recurrence. Recently, single cell sequencing studies of primary GBM have defined various tumor states, specifically the neural progenitor-like (NPC-like), oligodendrocyte progenitor (OPC-like), astrocyte-like(AC-like), and mesenchymal(MES-like) states. Yet, role these states play in treatment resistance recurrence not well defined. Furthermore, spatial microenvironmental (TME) context cells clearly plays response....
For training an encoder network to perform amortized variational inference, the Kullback-Leibler (KL) divergence from exact posterior its approximation, known as inclusive or forward KL, is increasingly popular choice of objective due mass-covering property minimizer. However, minimizing this challenging. A existing approach, Reweighted Wake-Sleep (RWS), suffers heavily biased gradients and a circular pathology that results in highly concentrated distributions. As alternative, we propose...
Abstract Background A folding RNA molecule encounters multiple opportunities to form non-native yet energetically favorable pairings of nucleotide sequences. Given this forbidding free-energy landscape, mechanisms have evolved that contribute a directed and efficient process, including catalytic proteins error-detecting chaperones. Among structural molecules we make distinction between “bound” molecules, which are active as part ribonucleoprotein (RNP) complexes, “unbound,” with...
Abstract This paper studies measurement linkage . An example from cell biology helps explain the problem: imagine for a given we can either sequence cell’s RNA or examine its morphology, but not both. Given what do expect to see in RNA? RNA, morphology? More broadly, of one type, predict measurements other type? problem arises many scientific and technological fields. To solve this problem, develop nonparametric approach dub “Markov link method” (MLM). The MLM makes conditional independence...
We consider an n-dimensional Brownian Motion trapped inside a bounded convex set by normally-reflecting boundaries. It is well-known that this process uniformly ergodic. However, the rates of ergodicity are not well-understood, especially in regime very high-dimensional sets. Here we present new bounds on these for sets with given diameter. Our do depend upon smoothness boundary nor value ambient dimension, n.
A reversible diffusion process is initialized at position ${x}_{0}$ and run until it first hits any of several targets. What the probability that terminates a particular target? We propose computationally efficient approach for estimating this probability, focused on those situations in which takes long time to hit target. In these cases, direct simulation hitting probabilities becomes prohibitively expensive. On other hand, if timescales are sufficiently long, then system will essentially...
Gaussian Processes (GPs) provide powerful probabilistic frameworks for interpolation, forecasting, and smoothing, but have been hampered by computational scaling issues. Here we investigate data sampled on one dimension (e.g., a scalar or vector time series at arbitrarily-spaced intervals), which state-space models are popular due to their linearly-scaling costs. It has long conjectured that general, able approximate any one-dimensional GP. We the first general proof of this conjecture,...
We consider the problem of multiple hypothesis testing when there is a logical nested structure to hypotheses. When one inside another, outer must be false if inner false. model as directed acyclic graph, including chain and tree graphs special cases. Each node in graph rejecting requires also all its ancestors. propose general framework for adjusting node-level test statistics using known constraints. Within this framework, we study smoothing procedure that combines each with descendants...
Amortized variational inference produces a posterior approximation that can be rapidly computed given any new observation. Unfortunately, there are few guarantees about the quality of these approximate posteriors. We propose Conformalized Neural Variational Inference (CANVI), procedure is scalable, easily implemented, and provides guaranteed marginal coverage. Given collection candidate amortized approximators, CANVI constructs conformalized predictors based on each candidate, compares using...
Modern cell-perturbation experiments expose cells to panels of hundreds stimuli, such as cytokines or CRISPR guides that perform gene knockouts. These are designed investigate whether a particular is upregulated downregulated by exposure each treatment. However, due high levels experimental noise, typical estimators up- down-regulated make many errors. In this paper, we two contributions. Our first contribution new estimator regulatory effect makes use Gaussian processes and factor analysis...
Abstract Glioblastoma (GBM), an aggressive, primary tumor is composed of multiple clonal populations. Resistance to therapy a hallmark GBM, making it deadly disease. Upon therapeutic stress, heterogeneity and glial stem cell plasticity within the contributes reprogramming transcriptional activity. This alters landscape generates resistant Single RNA sequencing studies GBM have defined various states namely, NPC-like (Neuronal precursor-like), OPC-like (oligodendrocyte precursor-like)...
Abstract Background A folding RNA molecule encounters multiple opportunities to form non-native yet energetically favorable pairings of nucleotide sequences. Given this forbidding free-energy landscape, mechanisms have evolved that contribute a directed and efficient process, including catalytic proteins error-detecting chaperones. Among structural molecules we make distinction between “bound” molecules, which are active as part ribonucleoprotein (RNP) complexes, “unbound,” with...