Quantification of Osteoclasts in Culture, Powered by Machine Learning.

In vitro osteoclastogenesis is a central assay in bone biology to study the effect of genetic and pharmacologic cues on the differentiation of bone resorbing osteoclasts. To date, identification of TRAP+ multinucleated cells and measurements of osteoclast number and surface rely on a manual tracing requiring specially trained lab personnel. This task is tedious, time-consuming, and prone to operator bias. Here, we propose to replace this laborious manual task with a completely automatic process using algorithms developed for computer vision. To this end, we manually annotated full cultures by contouring each cell, and trained a machine learning algorithm to detect and classify cells into preosteoclast (TRAP+ cells with 1-2 nuclei), osteoclast type I (cells with more than 3 nuclei and less than 15 nuclei), and osteoclast type II (cells with more than 15 nuclei). The training usually requires thousands of annotated samples and we developed an approach to minimize this requirement. Our novel strategy was to train the algorithm by working at "patch-level" instead of on the full culture, thus amplifying by >20-fold the number of patches to train on. To assess the accuracy of our algorithm, we asked whether our model measures osteoclast number and area at least as well as any two trained human annotators. The results indicated that for osteoclast type I cells, our new model achieves a Pearson correlation (r) of 0.916 to 0.951 with human annotators in the estimation of osteoclast number, and 0.773 to 0.879 for estimating the osteoclast area. Because the correlation between 3 different trained annotators ranged between 0.948 and 0.958 for the cell count and between 0.915 and 0.936 for the area, we can conclude that our trained model is in good agreement with trained lab personnel, with a correlation that is similar to inter-annotator correlation. Automation of osteoclast culture quantification is a useful labor-saving and unbiased technique, and we suggest that a similar machine-learning approach may prove beneficial for other morphometrical analyses.

Rich cell-type-specific network topology in neocortical microcircuitry.

Uncovering structural regularities and architectural topologies of cortical circuitry is vital for understanding neural computations. Recently, an experimentally constrained algorithm generated a dense network reconstruction of a ∼0.3-mm3 volume from juvenile rat somatosensory neocortex, comprising ∼31,000 cells and ∼36 million synapses. Using this reconstruction, we found a small-world topology with an average of 2.5 synapses separating any two cells and multiple cell-type-specific wiring features. Amounts of excitatory and inhibitory innervations varied across cells, yet pyramidal neurons maintained relatively constant excitation/inhibition ratios. The circuit contained highly connected hub neurons belonging to a small subset of cell types and forming an interconnected cell-type-specific rich club. Certain three-neuron motifs were overrepresented, matching recent experimental results. Cell-type-specific network properties were even more striking when synaptic strength and sign were considered in generating a functional topology. Our systematic approach enables interpretation of microconnectomics 'big data' and provides several experimentally testable predictions.

Past-future information bottleneck in dynamical systems.

Biological systems need to process information in real time and must trade off accuracy of presentation and coding costs. Here we operationalize this trade-off and develop an information-theoretic framework that selectively extracts information of the input past that is predictive about the output future, obtaining a generalized eigenvalue problem. Thereby, we unravel the input history in terms of structural phase transitions corresponding to additional dimensions of a state space. We elucidate the relation to canonical correlation analysis and give a numerical example. Altogether, this work relates information-theoretic optimization to the joint problem of system identification and model reduction.

The minimum information principle and its application to neural code analysis.

The study of complex information processing systems requires appropriate theoretical tools to help unravel their underlying design principles. Information theory is one such tool, and has been utilized extensively in the study of the neural code. Although much progress has been made in information theoretic methodology, there is still no satisfying answer to the question: "What is the information that a given property of the neural population activity (e.g., the responses of single cells within the population) carries about a set of stimuli?" Here, we answer such questions via the minimum mutual information (MinMI) principle. We quantify the information in any statistical property of the neural response by considering all hypothetical neuronal populations that have the given property and finding the one that contains the minimum information about the stimuli. All systems with higher information values necessarily contain additional information processing mechanisms and, thus, the minimum captures the information related to the given property alone. MinMI may be used to measure information in properties of the neural response, such as that conveyed by responses of small subsets of cells (e.g., singles or pairs) in a large population and cooperative effects between subunits in networks. We show how the framework can be used to study neural coding in large populations and to reveal properties that are not discovered by other information theoretic methods.

Correlations between groups of premotor neurons carry information about prehension.

How distinct parameters are bound together in brain activity is unknown. Combination coding by interneuronal interactions is one possibility, but, to coordinate parameters, interactions between neuronal pairs must carry information about them. To address this issue, we recorded neural activity from multiple sites in the premotor cortices of monkeys that memorized reach direction and grasp type followed by actual prehension. We found that correlations between individual spiking neurons are generally weak and carry little information about prehension. In contrast, correlations and synchronous interactions between small groups of neurons, quantified by multiunit activity (MUA), are an order of magnitude stronger. A substantial fraction of the information carried by pairwise interactions between MUAs is about combinations of reach and grasp. This contrasts with the information carried by individual neurons and individual MUAs, which is mainly about reach and/or grasp but much less about their combinations. The main contribution of pairwise interactions to the coding of reach-grasp combinations is when animals memorize prehension parameters, consistent with an internal composite representation. The informative interactions between neuronal groups may facilitate the coordination of reach and grasp into coherent prehension.