TopMSV: A Web-Based Tool for Top-Down Mass Spectrometry Data Visualization

Abstract

Top-down mass spectrometry (MS) investigates intact proteoforms for proteoform identification, characterization, and quantification. Data visualization plays an essential role in top-down MS data analysis because proteoform identification and characterization often involve manual data inspection to determine the molecular masses of highly charged ions and validate unexpected alterations in identified proteoforms. While many software tools have been developed for MS data visualization, there is still a lack of web-based visualization software designed for top-down MS. Here we present TopMSV, a web-based tool for top-down MS data processing and visualization. TopMSV provides interactive views of top-down MS data using a web browser. It integrates software tools for spectral deconvolution and proteoform identification and uses analysis results of the tools to annotate top-down MS data.

1. Introduction

Top-down mass spectrometry (MS) is a commonly used method for proteoform identification, characterization, and quantification^{1, 2}. Top-down MS can identify various proteoforms of a protein with combinatorial patterns of post-translational modifications (PTMs)^3-5 and has found many applications in studying proteoform function and diseases ^{6, 7}. In recent years, advances in instrumentation and bioinformatics tools have further boosted the popularity of top-down MS.

Data visualization plays an essential role in top-down MS data analysis. Because proteoforms identified by top-down MS often contain PTMs and unexpected alterations^{3, 8}, visualization tools are used to characterize these proteoforms by manual inspection of proteoform-spectrum-matches (PrSMs). Visualization tools are also routinely used to determine the molecular masses of highly charged ions in top-down mass spectra.

Software tools for MS data visualization^9-12 are either standalone or web-based. Standalone tools, such as BatMass⁹, TOPPView¹⁰, PDV¹³, and pLabel¹⁴, offer graphic user interfaces to visualize MS1 and tandem mass spectrometry (MS/MS) data as well as spectral identifications. Moreover, PDV¹³ supports proteogenomics data in the proBAM and proBed format and proteomics data in the PRIDE XML and PeptideAtlas sptxt format. TopPIC¹⁵, ProSight Lite¹¹, and LcMsSpectator¹⁶ provide functions to visualize proteoform identifications reported by top-down MS data analysis. Software packages like ProSightPC¹⁷ and MASH Explorer¹⁸ offer an integrated environment for top-down MS data analysis and visualization.

Compared with standalone tools, web-based visualization tools are easier to install, maintain, and access. xiSPEC¹⁹ and IPSA²⁰ are web-based tools for users to upload and browse MS data. xiSPEC supports the visualization of MS/MS spectra and identified peptides, especially cross-linked peptides. IPSA generates 2-dimensional (2D) interactive views of MS/MS spectra and their annotations reported by spectral identification tools. JS-MS¹² is a web-based tool for interactive 2D and 3-dimensional (3D) visualization of liquid chromatography-mass spectrometry (LC/MS) data. It parses mzML data and plots peaks in the web-browser and has various functions such as 2D views, bookmarks, and manual annotation of mass spectra.

These web-based tools can be utilized to visualize top-down MS data, but they are not designed specifically for top-down MS. They often lack functions to integrate proteoform identifications in top-down MS, which tend to be longer and contain more alterations than peptide identifications in bottom-up MS. Also, top-down MS data have many highly charged isotopic peaks of proteoforms and fragments, so we need functions for annotating these peaks in top-down MS data visualization.

Here we present TopMSV, a web-based tool for top-down MS data processing and visualization. It provides software tools for top-down spectral deconvolution and proteoform identification by database search as well as various 2D and 3D views for top-down MS data visualization. Researchers can use TopMSV to quickly explore LC profiles, LC-MS maps, MS and MS/MS spectra, and spectral deconvolution and proteoform identification results. Developers can also use JavaScript libraries provided by TopMSV to implement their interfaces for MS data visualization.

2. Methods

The server-side of TopMSV was developed using Node.js, Express.js, and C++. Node.js is an open-source, cross-platform, JavaScript runtime environment, and Express.js is an open-source web application framework for Node.js. The tools for MS data format conversion were implemented in C++.

2.1. Architecture of the server

The server side of TopMSV consists of two layers: a storage layer and an application layer (Fig. 1). In the storage layer, TopMSV combines file folders and SQLite databases to store MS data. TopMSV creates a file folder for each uploaded MS data file, and all databases and other related files are stored in the folder. Because spectra and peaks in MS data files are frequently queried in data visualization, several SQLite database tables are created for each MS data file to speed up queries. An SQLite server database is designed to store the information of the server and the MS data files. Two libraries are used for database operations: sqlite3 and better-sqlite3. Sqlite3 is a native SQLite module in Node.js supporting full functions for SQL queries, but it is inefficient for synchronous queries. Because the synchronous mode of better-sqlite3 is faster than sqlite3, it is used for queries that need quick responses.

Figure 1.

The architecture of the web-server of TopMSV. The server side (back end) consists of an application layer and a storage layer, and the browser side (front end) has a view layer. The application layer functions are implemented in Node.js or C++, and a task scheduler is used to execute the C++ tools. In the storage layer, MS data and analysis results are stored in SQLite databases, XML files, and text files; the server setting and project information are stored in an SQLite database (Server database). The browser-side functions are implemented in JavaScript.

The application layer connects the web-browser with the storage layer and contains main functions for data processing and querying. The Node.js functions are divided into two types. The first type is data management functions for receiving uploaded data files from the browser side, storing data into databases, and sending data in the JSON format to the browser side for visualization. For instance, a user can retrieve a spectral peak list from an MS database by a scan number.

The second type is a task scheduler. It balances submitted tasks and available resources in the server to handle requests from clients without disruption. The task scheduler automatically detects CPU resources and maintains a task queue at the server. After initialization, the task scheduler checks the queue every second. If there are available resources and remaining tasks on the waiting list, the task scheduler executes the next task in the queue.

The application layer contains C++ tools for MS data file format conversion, spectral deconvolution, and database search. TopMSV uses a C++ function to read mzML files and store MS data into SQLite databases and incorporates TopFD²¹ and TopPIC¹⁵ for spectral deconvolution and proteoform identification by database search. TopFD and TopPIC report analysis results in text and XML files. A Python tool based on the psims library²² was also developed to convert text identification files reported by TopPIC into mzIdentML files. TopMSV uses Node.js functions to read these files and store the results into SQLite databases. Because it is time-consuming to analyze large MS data using the C++ tools, TopMSV sends users email notifications containing job statuses and links of results after data analysis jobs are completed.

2.2. Database design

We create an SQLite database for each MS data file uploaded to the server. The spectral information and peaks in an MS data file are stored in three tables: SPECTRA, PEAKS, and SCAN_PAIRS, where SCAN_PAIRS contains the MS1 scan of each MS/MS scan. We also use database tables to store spectral deconvolution and spectral identification results reported by TopFD and TopPIC.

2.2.1. Multiple-layer storage of MS1 data

A view of all MS1 spectral peaks is provided to explore proteoform features and abundances in LC-MS data. The number of all MS1 peaks is often very large, making it challenging to visualize them in a web browser. For example, a top-down MS data file of a proteome-wide study of human cells often contains millions of MS1 peaks. Several methods have been designed to address this problem¹². We use a multiple-layer data structure to store MS1 peaks in SQLite tables to speed up the queries and visualization of MS1 peaks.

We use a 2D grid on the LC-MS map to generate a database table for all MS1 peaks. Let (p,q) be the grid's cell size, where p is the m/z bin width and q is the retention time bin width. The cell in the ith column and jth row of the grid contains the peaks with an m/z value in [(i-1)p, ip] and a retention time in [(j-1)q, jq]. For example, when the m/z bin width is 1 m/z, the retention time bin width is 1 second, the whole m/z range of the MS data is 0 to 2000 m/z, and the entire retention time range is 0 to 1000 seconds, there will be 2000 × 1000 cells in the grid. The peak with the highest intensity in a cell is called its representative peak. The database table for the grid contains all the representative peaks and no other peaks. An index based on the m/z value and retention time is also created to speed up queries of peaks.

There is a tradeoff between the speed and data completeness in selecting the cell size for generating database tables of peaks. When a small cell size is chosen, most peaks are stored in the table, and the data completeness is improved, but the speed becomes slow when peaks in a large window (a large m/z range and a large retention time range) need to be shown. When a large cell size is chosen, a small number of peaks are stored, and the speed is fast, but many peaks are missing when peaks in a small window need to be shown. To address this problem, we generate multiple database tables using 2D grids with various cell sizes. These tables contain different numbers of peaks. To show peaks in a large window, we select peaks from a table with a large cell size to reduce reported peaks. Because representative peaks are the highest in cells, the reported peaks are incomplete but still provide a good summary of all peaks. To show peaks in a small window, we choose peaks from a table with a small cell size to obtain most of the peaks in the window.

The cell sizes for generating database tables are selected as follows. The first cell size (a₁, b₁) is chosen to be small enough so that each cell contains at most one peak. Its corresponding table contains all peaks in the data file. The second cell size (a₂, b₂) is a parameter of the server. For k ≥ 2 the m/z and retention time bin widths of the k+1th cell size (a_k+1, b_k+1) are two times those of the kth cell size (a_k, b_k). That is, a_k+1 = 2a_k and b_k+1=2b_k. For example, when the second cell size is 0.1 m/z and 0.1 seconds, the third cell size is 0.2 m/z and 0.2 seconds. When the cell size increases, the database table size decreases. We will stop adding new tables when the database table size is below a threshold (3000 peaks). Although many tables are used for storing peaks, the total number of peaks in the tables is not significantly increased. Let n_k be the number of peaks in the kth table. When $n_{k+1}\le \frac{n_k}{2}$ for k ≥ 1, the total number of peaks in the tables is less than two times the data's peak number.

2.2.2. Peak queries for LC-MS views

Here we present the method for choosing a cell size and its database table based on the m/z and retention time ranges of an LC-MS data visualization window. Let (x,y) be the size of a visualization window, where x is the m/z range and y is the retention time range. Let M be the number of peaks we expect to show in a window, which is a server parameter. First, we find the ratio between the visualization window and the window containing the whole data. Let X be the maximum m/z value and Y the maximum retention time in the data. The ratio is calculated as $r=\frac{xy}{XY}$. Next, we estimate the number of peaks in the window for each database table. For a database table with N peaks, the estimated number of peaks in the window is rN. Third, we choose the table so that the value rN is the closest to M.

Using multiple tables and the table selection method, we can obtain similar numbers of peaks from LC-MS data for visualization no matter the window size is large or small. When the number of peaks reported by a query is much larger than M, we will sort the peaks by their intensities in the decreasing order and keep only the top M ones.

2.3. Web browser-based visualization

TopMSV provides various views for LC/MS and MS/MS data as well as their spectral deconvolution and proteoform identification results. 2D views of mass spectra and identified PrSMs are implemented using the D3.js library, and 3D views of LC-MS data using the three.js library. Users can also upload MS data and submit data analysis tasks (Fig. S1 and S2 in the supplementary material).

2.3.1. Views of LC-MS data

Two views are used for the visualization of LC-MS data. The first view (Fig. 2) shows the liquid chromatogram of the data. The x-axis of the view is the retention time in minutes, and the y-axis is the normalized sum of the peak intensities of detected precursor ions (range from 0% to 100%). There is also a vertical line in the view showing the retention time selected by the user to display MS1 and MS/MS spectra.

Figure 2.

An example of the liquid chromatogram view of LC-MS data.

The second view has three dimensions: x: m/z value, y: retention time, and z: peak intensity (Fig. 3). In this view, each peak is represented by a line. We assign colors to peaks based on their intensities to help users distinguish between high and low ones. For a peak with an intensity h, the normalized intensity is $\frac{\log h}{\log H}$, where H is the highest peak intensity in the data. The color of the peak is assigned using the normalized intensity and a color scheme in BatMass⁹. Blue or yellow color represents low-intensity peaks; red or orange color represents high-intensity ones.

Figure 3.

An example of the 3D view of LC-MS data. Each peak is represented by a line, and its peak intensity determines its color: orange or red for high-intensity peaks and yellow or blue for low-intensity ones. The peaks in pink are from a spectrum chosen by the user to inspect.

The view provides drag, zoom, and rotation functions as well as peak intensity scaling functions for the user to explore the data. When the 3D view is rotated so that the viewer faces the direction of the negative z-axis, it becomes a 2D view, which is easier for the user to inspect isotopic patterns of precursor ions (Fig. 4). The user can highlight peaks of one MS1 scan. Top-down spectral deconvolution tools^{16, 21, 23} report proteoform features in LC-MS data. A proteoform feature is a group of isotopic peaks of a proteoform with a charge state in the LC-MS map, resulting from molecular isotopes of the proteoform²⁴. A proteoform may have several proteoform features with various charge states in the LC-MS map. Spectral deconvolution tools report the monoisotopic m/z value, charge, range of m/z values, and range of the retention time of a proteoform feature. TopMSV stores proteoform features using the feature file format (CSV files) in BatMass⁹. Each feature is represented by an m/z value range (lowMz, highMz) and a retention time range (lowRT, highRT), and annotated by a rectangle of dotted lines defined by the two ranges in the LC-MS map.

Figure 4.

An example of the 2D view of LC-MS data. The peaks in pink are from a spectrum chosen by the user to inspect.

2.3.2. Views of spectra

Two views are used to show MS1 and MS/MS spectra as well as their spectral deconvolution and database search results. The first view shows the peaks and isotopic envelopes reported by spectral deconvolution in a mass spectrum (Fig. 5). Users can change the m/z range of the view and scale peak intensities by drag and zoom functions. Spectral deconvolution reports many isotopic envelopes, each of which contains peaks from the same ion with various isotopic compositions. For each experimental isotopic envelope, we use the Averagine model²⁵ to compute its theoretical isotopic distribution, shown by colored dots. Users can easily evaluate reported isotopic envelopes by comparing the peaks in an experimental isotopic envelope and the dots in the theoretical distribution,

Figure 5.

An example of the view of a pair of MS1 and MS/MS spectra. The top is a view of the m/z range [808, 815] of an MS1 spectrum, in which the theoretical isotopic distributions of two precursor ions are shown by blue and red dots, respectively. The bottom is a view of a whole MS/MS spectrum.

The second view shows deconvoluted monoisotopic masses in an MS/MS spectrum (Fig. 6). For each isotopic envelope reported by spectral deconvolution, we compute its monoisotopic mass and use the sum of the peak intensities in the theoretical distribution as the intensity of the monoisotopic mass. The view consists of three components: the identified proteoform sequence and their theoretical fragment masses at the top, the monoisotopic masses in the middle, and the m/z errors between matched theoretical and experimental masses at the bottom.

Figure 6.

An example of the view of monoisotopic masses of MS/MS spectra. The proteoform sequence matched to the spectrum and its theoretical fragment masses are shown at the top, monoisotopic masses and their annotations in the middle, and an error plot of matched theoretical and experimental masses at the bottom.

2.3.3. A view of proteoforms

TopMSV offers a view of proteoforms identified by top-down mass spectra (Fig. 7), in which all matched theoretical fragment masses of the proteoform are annotated. Modified amino acid residues are also highlighted: amino acids with fixed modifications are shown in red, and candidate amino acid residues for unexpected mass shifts are annotated with a blue background. Moreover, TopMSV provides a table of monoisotopic masses with their matched theoretical masses and functions for adding and removing modifications of the proteoform, helping users characterize the proteoform with manual inspection.

Figure 7.

An example of the view of proteoforms. Amino acids with fixed modification are shown in red, and those with unexpected mass shifts with a blue background.

3. Results

We tested TopMSV using a top-down LC-MS/MS data set of a breast cancer xenograft sample. TopMSV was installed on a web server with an Intel Xeon Silver 4208 2.10GHz CPU and 64GB memory. In the LC-MS/MS experiment, the protein extracted from the sample was analyzed by a 180 min LC-MS/MS experiment, in which a Waters nanoACQUITY UPLC system (Milford, MA) and a Velos Orbitrap Elite mass spectrometer (Thermo Fisher, San Jose, CA) were used. Precursor spectra were collected using a resolution of 240,000 and MS/MS spectra using a resolution of 120,000. Ten replicates of LC-MS/MS data were generated from the sample. More details of the experiment can be found in Ref. 16. We used the fourth replicate in the test, which contained 1148 MS1 spectra and 6888 MS/MS spectra.

3.1. Database size and speed

The raw file of the data was converted to an mzML file using msconvert in ProteoWizard ²⁶. The size of the mzML file was 365 MB. After the mzML file was uploaded to the server, the running time for storing the peak information into an SQLite database was 10.6 minutes. The size of the database was 4.3 GB.

We tested the speed of the 3D view for data queries from the database and data rendering using three.js with four window sizes (5 m/z, 0.5 min), (50 m/z, 5 mins), (500 m/z, 50 min), and (1000 m/z, 100 min). For each window size, we generated 1000 pairs of random m/z and retention time ranges. For the four window sizes, the average database query times were 31.7, 41.7, 109.0, and 138.4 milliseconds; the average times for plotting and rendering the peaks were 15.3, 34.9, 192.2, 280.4 milliseconds, and the average numbers of peaks shown in the window were 1, 166, 1277, and 1995 respectively. The user can query the data using various window sizes without delays because of the fast response time.

4. Conclusions

TopMSV is a web-based tool for visualizing LC/MS and MS/MS data. It provides interactive 2D and 3D views of MS data and functions for top-down MS spectral deconvolution and identification. Because it processes and stores MS data at the web-server, the user needs only a web browser to use the full functionality of TopMSV, simplifying the installation and maintenance of the tool. The user can submit data sets to the web server for private data visualization and public data sharing. The various views and tables provided by TopMSV enable the user to manually inspect MS data for proteoform identification, characterization, and quantification. TopMSV supports only the Google Chrome web browser. In the future, we will expand the support for other web browsers like Firefox and add new functions, such as comparing two or more data sets.