Biology subjects

Denk, W.

Publications and source records attributed to Denk, W..

4 recordsLinked to original sources

bioRxiv2023

Efficient Search for Circuit Structure by 'Smooth-Index' Matrix-Reordering

Volume electron microscopy together with computer-based image analysis are yielding neural circuit diagrams of ever larger regions of the brain [1-10]. These datasets are usually represented in a cell-to-cell connectivity matrix and contain important information about prevalent circuit motifs allowing to directly test various theories on the computation in that brain structure [11,12]. Of particular interest are the detection of cell assemblies and the quantification of feedback, which can profoundly change circuit properties. While the ordering of cells along the rows and columns doesnt change the connectivity, it can make special connectivity patterns recognizable. For example, ordering the cells along the flow of information, feedback and feedforward connections are segregated above and below the main matrix diagonal, respectively. Different algorithms are used to renumber matrices such as to minimize a given cost function, but either their performance becomes unsatisfying at a given size of the circuit or the CPU time needed to compute them scales in an unfavorable way with increasing number of neurons [13-15]. Based on previous ideas [16-18], we describe an algorithm which is effective in matrix reordering with respect to both its performance as well as to its scaling in computing time. Rather than trying to reorder the matrix in discrete steps, the algorithm transiently assigns a real-valued parameter to each cell describing its location on a continuous axis ( smooth-index) and finds the parameter set that minimizes the cost. We find that the smooth-index algorithm outperforms all algorithms we compared it to, including those based on topological sorting. Author SummaryConnectomic data provide researchers with neural circuit diagrams of ever larger regions of the brain. These datasets are usually represented in a cell-to-cell connectivity matrix and contain important information about prevalent circuit motifs. Such motifs, however, only become visible if the connectivity matrix is reordered appropriately. For example, ordering the cells along the flow of information, feedback and feedforward connections are segregated above and below the main matrix diagonal, respectively. While most previous approaches rely on topological sorting, our method treats the discrete vertex indices as real numbers ( smooth-index) along independent parameter axes and defines a differentiable cost function, thus, allowing gradient-based algorithms to find a minimum. The parameter set at this minimum is then re-discretized to reorder the connectivity matrix accordingly. We find our method to scale favorably with the circuit size and to outperform all algorithms we compared it to.

neuroscience↗

bioRxiv2020

Label-free single-instance protein detection in vitrified cells

A general method to map molecular interactions and conformational states in structurally intact cells would find wide application in biochemistry and cell biology. We used a library of images-- calculated on the basis of known structural data--as search templates to detect targets as small as the "head" domain (350 kDa) of the ribosomes small subunit in single-tilt electron cryo-micrographs by cellular high resolution template matching (cHRTM). Atomically precise position and orientation estimates reveal the conformation of individual ribosomes and enable the detection of specifically bound ligands down to 24 kDa. We show that highly head-swivelled states are likely to play a role in mRNA translocation in living cells. cHRTM outperforms cryo-electron tomography three-fold in sensitivity and completely avoids the vicissitudes of exogenous labelling.

biophysics↗

bioRxiv2020

An anatomical substrate of credit assignment in reinforcement learning

A key problem in learning is credit assignment. Biological systems lack a plausible mechanism to implement the backpropagation approach, a method that underlies much of the dramatic progress in artificial intelligence. Here, we use automated connectomic analysis to show that the synaptic architecture of songbird basal ganglia (Area X) supports local credit assignment using a variant of a node perturbation algorithm proposed in a model of reinforcement learning. Using two volume electron microscopy (vEM) datasets, we find that key predictions of the model hold true: axons that encode exploratory variability terminate predominantly on dendritic shafts, while axons that encode song timing (context) terminate predominantly on spines. Based on the detailed EM data, we then built a biophysical model of reinforcement learning that suggests that the synaptic dichotomy between variability and context encoding axons facilitates efficient learning. In combination, these findings provide strong evidence for a general, biologically plausible credit assignment model in vertebrate basal ganglia learning. One Sentence SummaryUsing automated connectomic analysis and biophysical modeling, we show how the basal ganglia could solve the credit assignment problem on the synaptic level.

neuroscience↗

bioRxiv2017

High-Precision Automated Reconstruction of Neurons with Flood-filling Networks

Reconstruction of neural circuits from volume electron microscopy data requires the tracing of complete cells including all their neurites. Automated approaches have been developed to perform the tracing, but without costly human proofreading their error rates are too high to obtain reliable circuit diagrams. We present a method for automated segmentation that, like the majority of previous efforts, employs convolutional neural networks, but contains in addition a recurrent pathway that allows the iterative optimization and extension of the reconstructed shape of individual neural processes. We used this technique, which we call flood-filling networks, to trace neurons in a data set obtained by serial block-face electron microscopy from a male zebra finch brain. Our method achieved a mean error-free neurite path length of 1.1 mm, an order of magnitude better than previously published approaches applied to the same dataset. Only 4 mergers were observed in a neurite test set of 97 mm path length.

neuroscience↗