PMAnalyzer: a new web interface for bacterial growth curve analysis

Abstract Summary Bacterial growth curves are essential representations for characterizing bacteria metabolism within a variety of media compositions. Using high-throughput, spectrophotometers capable of processing tens of 96-well plates, quantitative phenotypic information can be easily integrated into the current data structures that describe a bacterial organism. The PMAnalyzer pipeline performs a growth curve analysis to parameterize the unique features occurring within microtiter wells containing specific growth media sources. We have expanded the pipeline capabilities and provide a user-friendly, online implementation of this automated pipeline. PMAnalyzer version 2.0 provides fast automatic growth curve parameter analysis, growth identification and high resolution figures of sample-replicate growth curves and several statistical analyses. Availability and Implementation PMAnalyzer v2.0 can be found at https://edwards.sdsu.edu/pmanalyzer/. Source code for the pipeline can be found on GitHub at https://github.com/dacuevas/PMAnalyzer. Source code for the online implementation can be found on GitHub at https://github.com/dacuevas/PMAnalyzerWeb. Supplementary information Supplementary data are available at Bioinformatics online.


Introduction
Bacteria growth curves have been proven useful in several studies where cellular proliferation is the primary measured response; for example, in characterizing novel bacteriophage proteins (Sanchez et al., 2015), in describing mutant bacterial strains (Perkins and Nicholson, 2008) and in elucidating genome-scale metabolic models (Cuevas et al., 2016a,b;Kim and Reed, 2014). Automated multi-96well plate spectrophotometer systems provide the ability to test hundreds of these hypotheses at a time, monitoring biomass accumulation over a span of hours or days.
Downstream analytical tools need to (i) easily and quickly process datasets in various formats, (ii) provide transparency and clear explanations of their pipeline and (iii) produce results flexible enough to become integrated with other quantitative information.
Software capable of decomposing growth experimental data into usable metrics, such as the carrying capacity of an organism expressed in various growth media, is not freely and easily accessible. Specifically with 96-well plates, this type of software is normally developed for input of a single data format (e.g. GCAT (Bukhman et al., 2015)), to serve a specific spectrophotometer technology, or requires some programming knowledge to run and diagnose (Vehkala et al., 2015).
We present an improved version of our growth curve processing pipeline, PMAnalyzer v2.0, extended to include a variety of growth metrics, statistical results and a user-friendly online interface. The PMAnalyzer web tool allows users to analyze their growth curve data, view their results and export their information for subsequent analyses.

Bacterial growth curve model
Each bacterial growth curve is numerically fitted to the Zwietering logistic growth curve model (Zwietering et al., 1990) using a leastsquares fitting on the raw data: The model provides four key parameters that describe a growth curve: starting absorbance value y 0 (OD 600 nm), lag time k (hour), maximum growth rate l (OD 600 nm per hour) and biomass yield A (OD 600 nm). The parameter t is the time vector (hour). The result, b y, is a vector of logistic-fitted values. Each growth curve is modeled with this formula independent of technical and biological replicates. Afterwards, the four parameters are averaged between a sample's replicates to form a representative logistic-fitted growth curve of that sample. Raw growth curves, individual fitted growth curves, and representative fitted growth curves are all supplied in separate plain text files by the PMAnalyzer.

Growth analysis
The growth level metric, recapitulated from the previous version of PMAnalyzer, represents a single quantitative value of growth (Cuevas et al., 2016a). New metrics included in this updated version are: scaled growth level, growth class and mean squared error of each logistic fitting. Area under the curve calculations are also provided for raw data, normalized data, logistic fitted raw data and logistic fitted normalized data. For the complete list of the output metrics and their explanations, see the results documentation in the Supplementary Material.

Usage
The online implementation requires the user to input only a minimal amount of information. This updated version of PMAnalyzer is capable of processing datasets that have been generated in three unique formats, each defined from various projects using the pipeline. Each of those formats has been extensively described in PMAnalyzer's help webpage (https://edwards.sdsu.edu/pmanalyzer/help.html).
Most multi-plate spectrophotometers generate a separate data file of absorbance data for each plate. The user must uniquely name each file with a representative name and replicate identifier. This allows PMAnalyzer to perform various statistical analyses in regards to each sample and replicate.
Plate media information is optionally supplied to PMAnalyzer. The user has the option to either choose a pre-configured plate file (descriptions supplied in the help webpage), supply their own plate configuration file, or not use one at all. Without a plate configuration, PMAnalyzer differentiates environments based on the 96-well identifiers (e.g. A1, A2, B1, B2).

Job processing
During data processing, users are supplied a Job ID and a link to the log file. The Job ID is used to retrieve results in the future-this feature is supplied in the web tool, storing results for up to two weeks. The log file contains the verbose option output from the commandline. Here, users can view the current status of the job, what stage the job is at, and specific error messages, if any occur.

Results
At completion various tab-delimited plain text files are presented to be viewed or downloaded. Choosing to view a file opens a separate webpage where the contents of a file are presented in an HTML table for easy viewing.
Several results images are generated showing from both the raw and processed data (Fig. 1). If more than one sample is processed in a single job, comparative figures are generated displaying each sample's growth curves (averaged over a sample's replicates), heatmaps of growth levels (Fig. 1a) and density plots of each metric as a whole. In addition to these comparative figures, similar figures are generated for each individual sample (Fig. 1b, c).