A Systematic Evaluation of Interneuron Morphology Representations for Cell Type Discrimination

Abstract

Quantitative analysis of neuronal morphologies usually begins with choosing a particular feature representation in order to make individual morphologies amenable to standard statistics tools and machine learning algorithms. Many different feature representations have been suggested in the literature, ranging from density maps to intersection profiles, but they have never been compared side by side. Here we performed a systematic comparison of various representations, measuring how well they were able to capture the difference between known morphological cell types. For our benchmarking effort, we used several curated data sets consisting of mouse retinal bipolar cells and cortical inhibitory neurons. We found that the best performing feature representations were two-dimensional density maps, two-dimensional persistence images and morphometric statistics, which continued to perform well even when neurons were only partially traced. Combining these feature representations together led to further performance increases suggesting that they captured non-redundant information. The same representations performed well in an unsupervised setting, implying that they can be suitable for dimensionality reduction or clustering.

nGauge: Integrated and extensible neuron morphology analysis in Python

“…In addition to quantifying basic morphometric parameters, nGauge includes comprehensive utility functions for advanced neuroinformatics analysis (Table 1). For instance, nGauge implements the widely used Principal Component Analysis (PCA) to identify the differences between vectors of morphometrics (Gouwens et al, 2019;Laturnus, Kobak, et al, 2020). We performed PCA on a collection of pyramidal cells and basket cells (Miyamae et al, 2017) (Figure 5A) and a collection of tufted cells and mitral cells (Fukunaga et al, 2012) (Figure 5B).…”

“…Beyond single-value morphometrics, many tools have been integrated into nGauge for performing advanced analysis techniques. Influenced by recent work (Laturnus, Kobak, et al, 2020), nGauge includes tools to calculate 2D morphometric histograms; two example cells from (Miyamae et al, 2017) are shown (Figure 6). These plots can serve as "fingerprints" for the morphological properties of individual neurons.…”

“…However, all of these tools are either limited in features, or simply run L-Measure code in the background. The recent MorphoPy (Laturnus, von Daranyi, et al, 2020;Laturnus, Kobak, et al, 2020) package solves these problems by implementing many functions in native Python code, but has only limited ability to be extended to novel metric definitions, no standardized memory structure, and no ability to produce 3D visualizations. As a result, data integration for larger projects largely relies on bespoke methodologies which are limited in their reuse and accessibility.…”

Quantitative evaluation of neuron developmental morphology in vitro using the change-point test

“…Many different types of quantitative representations, such as density maps (Jefferis et al (2007); Laturnus, Kobak, and Berens (2020)), graph theory (Gillette and Grefenstette (2009); Heumann and Wittum (2009)), topology (Kanari et al (2018)), and morphometric statistics (Laturnus et al (2020); Polavaram, Gillette, Parekh, and Ascoli (2014); Uylings and van Pelt (2002)), have been applied to describe functionally different types of mature neurons. In addition, machine learning techniques also have been used for identifying neuron types (Laturnus et al (2020)) and for identifying neuronal polarity (Su et al (2021)). Laturnus et al (2020) noted the importance of the spatial extent and shape describing neuron connectivity in distinguishing cell types, instead of specific branching features.…”

“…In addition, machine learning techniques also have been used for identifying neuron types (Laturnus et al (2020)) and for identifying neuronal polarity (Su et al (2021)). Laturnus et al (2020) noted the importance of the spatial extent and shape describing neuron connectivity in distinguishing cell types, instead of specific branching features. Although these quantitative representations can characterize neuronal cell types, most have not been applied to discriminate between neurite growth stages or time points.…”