Interactively AUDIT Your Growth Curves with a Suite of R Packages

Abstract

Bottlenecks often occur during data analysis when studying microbial growth in liquid culture at large scale. A researcher can collect thousands of growth curves, repeated measures of a microbial liquid culture, at once in multiple micro titer plates by purpose-built robotic instruments. However, it can be difficult and time-consuming to inspect and analyze these data. This is especially true for researchers without programming experience. To enable this researcher, we created and describe an interactive application: Automated Usher for Data Inspection and Tidying (AUDIT). It allows the user to analyze growth curve data generated from one or more runs each with one or more micro titer plates alongside their experimental design. AUDIT covers input, pre-processing, summarizing, visual exploration and output. Compared to previously available tools AUDIT handles more data, provides live previews and is built from individually re-usable pieces distributed as R packages.

Keywords: chemical genetics; fitness; growth curve; yeast.

Figure 1

Overview Architecture and Conventions. A. AUDIT is an interactive, higher level, interface (i) atop a set of software modules/packages (ii) written and distributed via the underlying software platforms in R and JS (iii). The individual function groups (ii) are available as individual packages to enable re-use outside of the context of AUDIT. B. The stratification of individual measures across an experiment. $Experimen{t}_h$ contains the set of all runs from 1 to i ($Ru{n}_1,\dots ,Ru{n}_i$). Each Run contains the set of all plates from 1 to j ($P_1,\dots ,P_j$). Each plate contains the set of all wells from 1 to k ($W_1,\dots ,W_k$). Each well contains the set of measures from 1 to l ($M_1,\dots ,M_l$).

Figure 2

Automatic parsing and tidying of design and measures files uploaded to AUDIT. Screenshot of application’s input panel. Users upload one or more files via the file browser (arrow, top-left). Most inputs in AUDIT have a clickable ‘I’ that, when clicked, will display modal help box with additional information (arrow, second from top-left). After upload, files are automatically parsed and joined into runs. AUDIT logs parsing and joining actions for user review (arrow, bottom left). Alternatively, the user can choose example data to load by selecting one of the links provided. Users can choose seconds, minutes or hours for runtime units. An experimental summary table displays the resulting run(s) (arrow). For each run, the number of plates (n_plates), the number of wells per plate (n_wells) and the number of measures (n_measures) are reported. Selecting one or more rows of the experiment summary table will output a line plot for each well in the MTP’s physical layout. The limits of the x and y scales are per plate.

Figure 3

AUDIT reading modules transform varied raw input to consistent output. Examples of plain-text rendering of Bioscreen (top left), YG (top right), CG-12 (middle left) and GP1 (middle right) raw input files parsed to common tabular form. The diagram highlights optional fields with a question mark suffix. Following parsing of raw measures AUDIT joins experimental design data to the parsed measures, if available. Design files may be in generic CSV files or in plater format.

Figure 4

Live preview and flexible specification of varied preprocessing methods. Screenshot of preprocess panel. AUDIT makes six independent preprocessing options available (arrow, top-left). The Background Subtraction and Calibration input fields support arbitrary formula (code) execution (arrow, middle-right). The well preview selector table (arrow, top-right) contains data from selected runs in the experiment. When clicked, the ‘Preprocess Experiment’ button applies the preprocessing to the whole experiment (arrow, middle-left). To display a live plot preview, users select rows in the table. Then, AUDIT plots each well’s raw (solid line) and preprocessed measures (dashed line) (arrow, bottom-right).

Figure 5

Summarize growth curves by curve fit and group wells into reference and target sets. Screenshot of summarize panel in manual mode. AUDIT enables interactive specification of arbitrary growth models. Users specify a model via R’s formula interface. AUDIT estimates model coefficients for each curve, using non-linear least-squares optimization, from user-supplied initial estimates. When clicked, the ‘Summarize Experiment’ button applies the preprocessing to the whole experiment (arrow, middle-left). The well preview selector table (arrow, top-right) contains data from selected runs in the experiment. Selecting rows in the table, will cause a live line plot preview of each curve fit (solid line) atop the underlying measures data (points). Fit residual error is also plotted (red area segment). The application displays a tabular preview of the estimated summary metrics, including model specific coefficients, and the curve quality metrics.

Figure 6

Explore the distribution of summary metrics across an experiment and by group. Screenshot of AUDIT’s explore panel. The researcher chooses a summary metric and AUDIT creates a color map across the wells in an MTP view plot (arrow, top). Users can specify Reference-target groups interactively. AUDIT will generate line plots of the fitted growth curves along with a summary table of per-group means for each numeric summary metric computed.

Figure 7

Download tidied and preprocessed measures and summary data as CSV files. Screenshot of AUDIT’s output panel. Three data files are available for download: Measures data, containing tidied raw and preprocessed measures (left), summary-level data containing estimated metrics and model components/coefficients (middle), and summary-level fit quality data (right).

Figure 8

Case Study Overview. A. MTP view plot maps the values of the strain variable provided in the experimental design to the well fill color. A line plot of the measures data are plotted for each well. B. MTP view plot maps the drug identifier to the fill color and the dose in micromolar to the opacity of the rectangular wells. Line plots are according to the group specifications provided to the left of the plot. For each group, each reference curve is plotted in light, semi-transparent gray, in the target well while each the data for each target well is plotted in black. C. MTP view plot maps the values of the maximum growth rate reported in the summary metrics data. Key is provided on the right for each figure panel.