An Automated Respiratory Data Pipeline for Waveform Characteristic Analysis

Abstract

Whole body barometric flow through plethysmography is used to study respiratory output in mouse and rat genetic and disease models. Turn-key commercial and custom systems are both used to gather multiple data streams including respiratory-pressure waveforms, O2 and CO2 levels, etc. for a reasonable estimation of tidal volume and key metabolic parameters. These experiments typically produce large amounts of data that are time-consuming and difficult to organize and process. For analysis, the investigator may either choose from proprietary software with limited customization or use cumbersome, manual annotation subject to observer bias. Several software products may be combined to manage raw data manipulation, statistical analyses, and graphing. Commercial software is prohibitively expensive or bundled with turn-key systems, which limits options for those using bespoke respiratory measurement systems. To address these deficiencies, we developed a software pipeline for processing raw respiratory recordings and associated metadata (experimental design, age, weight, temperature, etc.) into operative respiratory outcomes, publication-worthy graphs, and robust statistical analyses. This pipeline consists of three modules. The first module is a Python-based graphical user interface (GUI) for uploading all relevant data files (raw waveforms, metadata, and configuration files) and defining both independent and dependent variables relevant to the experimental design, statistical analysis, and graphical outputs. Additionally, parameters can be programmed to define waveform feature segmentation to include manual annotations. In the second module, Breath-Caller, a Python program segments the respiratory waveform according to user-defined thresholds for movement artifacts and unique features, then quantifies respiratory cycle components and demarks unique features including sighs, apneas, expiratory reflexes, and user-defined entities. The module also allows for browsing the raw signals with markup of automatically annotated breaths to facilitate validation of breath detection and to permit manual annotation of features, if desired. The third module, SG-Runner, is an R-based program that accepts the Breath-Caller output along with configuration files from the GUI to graph the desired variables while performing a linear mixed effects statistical model analysis of the data with appropriate post-hoc tests. SG-Runner is soft-coded to dynamically adjust to the user settings defined in the GUI. Following graph generation, statistically significant differences in respiratory variables are automatically annotated in the graphs. Graphs are exported as individual SVG files for ease of editing and incorporation into figures and as a comprehensive R-markdown file for perusal and review in a web browser. The open-source program offers a facile and highly customizable platform for automated handling and analysis of animal respiratory and metabolic data that sets the stage for high-throughput studies and machine learning approaches on large data sets.