FastProtein—an automated software for in silico proteomic analysis

Introduction

The complexity of high-throughput sequencing data and the need for reproducible analysis are challenges that require integrated workflows (Wratten, Wilm & Göke, 2021). Some workflow managers for bioinformatics include community-driven projects and workflow management systems. Galaxy (The Galaxy Community et al.,, 2024) and nf-core (Ewels et al., 2020) are examples of community-driven projects, while Nextflow (Di Tommaso et al., 2017) and Snakemake (Mölder et al., 2021) are script-based workflow management systems. Workflow managers and automated software facilitate reproducible and scalable data analysis. However, Galaxy is the only platform with a user-friendly interface and a point-and-click feature to create workflows (Wratten, Wilm & Göke, 2021), while the remaining tools require the command-line interface.

Downstream analysis results in qualitative and quantitative features of proteins, which typically involve using several bioinformatics software packages in tandem but in a non-integrated workflow (Chen et al., 2020; Jiménez-Munguía et al., 2018). Although Blast2GO (Conesa et al., 2005) is an alternative tool widely used for functional annotation, it is closed sourced and requires a license. Proteomic analysis generates a considerable amount of computational data that require bioinformatics analysis (Vaudel et al., 2016).

FastProtein is an automated, user-friendly, and publicly available software that integrates functional annotation, database similarity search, and protein feature prediction to enable global proteomic profiling. Furthermore, the in silico results obtained through FastProtein can be used to characterize proteins of interest in search of biological insights.

Materials & Methods

Workflow

FastProtein uses a protein FASTA file as input to generate protein profiles. The workflow analysis begins by parsing and standardizing the input for different software (Fig. 1). Parsing and input validation are executed by BioJava (Lafita et al., 2019). Sequences with undetermined amino acids (represented by the letter “X”) are invalid and removed from the initial dataset before execution.

The first step of FastProtein involves biochemical feature prediction, which provides attributes such as protein length, molecular mass (kDa), hydropathy, isoelectric point (PI), and aromaticity (Lobry & Gautier, 1994). These processes are managed and executed through BioJava (Lafita et al., 2019). Subsequently, FastProtein identifies the N-glycosylation and endoplasmic reticulum retention domains using the PROSITE (Sigrist et al., 2013) database entries PS00001 and PS00014, respectively.

WoLF PSORT (Horton et al., 2007) is used to predict the subcellular locations of eukaryotic organisms. Transmembrane site, signal peptide, and glycosylphosphatidylinositol (GPI)-anchored predictions are performed using TMHMM-2 (Möller, Croning & Apweiler, 2001), SignalP 5.0, (Almagro Armenteros et al., 2019), and PredGPI (Pierleoni, Martelli & Casadio, 2008), respectively. Additionally, the transmembrane domain and signal peptide are predicted using Phobius (Käll, Krogh & Sonnhammer, 2004).

Functional annotations are optional and performed using InterProScan (Jones et al., 2014). The outputs are merged and parsed to obtain Pfam (Mistry et al., 2021) and PANTHER (Thomas et al., 2022) domains, InterPro (IPR) annotations (Jones et al., 2014), and gene ontology (GO) terms (Ashburner et al., 2000). The sets of GO terms associated with the protein are determined by analyzing all databases using InterProScan. These terms are organized into a file that can be imported into the WEGO 2.0 (Ye et al., 2018) platform for complementary analysis. This platform is used to group, visualize, compare, and generate GO plots.

GO terms provide quantitative reports on molecular functions, cellular components, and biological processes. This process generates a file containing the GO terms (one line per protein, followed by GO terms in table-separated files).

Another important step in this workflow is similarity analysis. FastProtein returns the best hit for each protein (with its identity and coverage percentage) using the BLASTp (Camacho et al., 2009) or DIAMOND (ultra-sensitive mode) (Buchfink, Reuter & Drost, 2021) algorithms. Local alignment using BLASTp is only available through the command-line interface (CLI). DIAMOND is the only local aligner available through the web server.

FastProtein provides six features to predict membrane protein: GPI-anchored, two predictions for transmembrane domains (TM, predicted via TMHMM-2.0; and PHOBIUS_TM, predicted via Phobius), subcellular localization predicted via WoLF PSORT (SL), GO, and IPR annotations. The presence of any one of these six features is sufficient to classify a protein as a membrane protein. Finally, a report file and a FASTA file are generated for proteins with membrane-related evidence.

User friendly web-based interface

A web module was developed using Python (v3.9.2) and Flask (v2.2.3) to execute FastProtein within a Docker container. This interface enables users submit new FastProtein executions and monitor the progress of their tasks (Fig. 2) and visualize results through charts (Fig. 3A) and an interactive table (Fig. 3B). Additionally, it includes modules for managing databases, users, and permissions.

Results

Only 124 proteins from the initial dataset were deemed eligible for analysis. Protein A0A1A8WBL6 had invalid sequences (represented by the letter “X”) and was removed from the dataset. The total analysis time for this dataset, using DIAMOND, was 2 min and 58 s, with 5 s exclusively required by FastProtein, and the remaining time by third-party software. The total execution time was 7 min and 13 s, using BLASTp, with 3 min and 49 s used for local alignment, 5 s by FastProtein, and the remaining by other third-party software.

The average molecular weight and isoelectric point in the P. malariae dataset were 98.29 ± 71.30 kDa and 7.79 ± 1.36, respectively. The distribution generated by FastProtein is shown in Fig. 3A. The average hydrophilicity and aromaticity were −0.50 ± 0.31 and 0.09 ± 0.03, respectively. Out of the analyzed proteins, 30 were predicted to contain transmembrane domains, three exhibited GPI anchoring, and an additional 30 were estimated to be membrane proteins. Furthermore, among the latter group, four exhibited multiple pieces of evidence supporting their localization to the membrane (GO, IPR, PHOBIUS_TM, and TM). Among the 124 P. malariae proteins, 11 were predicted to have signal peptides and 36 had ER retention domains, with the KEEL and KNEL domains being the most frequently occurring, existing in seven proteins each. Of the 124 proteins, 123 had N-glycosylation domains, with the NNS domain occurring most frequently (116 proteins). The subcellular locations are shown in Fig. 3A, wherein cytosol and nucleus are the most frequent ones, with 60 and 28 proteins, respectively.

For the proteome-wide analysis, the fastest runtime was 28.85 ± 5.74 min for the C. muris dataset (3,930 proteins; 3,924 processed and six ignored) using DIAMOND. The slowest runtime was 262.35 ± 2.63 min for the T. gondii dataset (8,404 proteins) using BLASTp. The execution time decreased to 77.75 ± 8.90 min with DIAMOND as the local aligner. Even though A. novofumigatus had 3,076 more proteins than T. gondii, it took 97.42 ± 0.21 min to analyze it using BLASTp,and 67.71 ± 1.73 min using DIAMOND. The best result in terms of proteins analyzed per minute was obtained for the A. novofumigatus dataset with 170 proteins (using DIAMOND), whereas the worst was obtained for T. gondii with 32 proteins (using BLASTp). For all DIAMOND executions, the alignment was completed within a few seconds. For BLASTp, the fastest execution required 8.02 ± 0.05 min and the slowest required 193.06 ± 0.92 min. Considering our entire workflow, the local alignment was the only step with different execution times, which is considerably similar to previously reported results using different alignment algorithms (Hernández-Salmerón & Moreno-Hagelsieb, 2020).

The average execution times for each software used in the pipeline were as follows: WoLF PSORT (∼2 min), TMHMM-2.0 (∼8 min), SignalP5 (∼2 min), PredGPI (∼2 min), Phobius (∼11 min), InterProScan (∼25 min), DIAMOND (<1 min), and BLASTp (∼52 min). The average time required for the internal execution of FastProtein, file generation, conversions, and calculations were approximately 1 min. Both files generated from the subset of P. malariae proteins and proteome-wide rounds are available at https://github.com/bioinformatics-ufsc/FastProtein (including the intermediate files, which were removed during processing). All data analyses are presented in Supplemental Information 3.

Discussion

FastProtein is a user-friendly and easy-to-install protein analysis pipeline tool that provides important information about protein datasets. FastProtein integrates calculations of molecular weight, isoelectric point, hydropathy, and aromaticity with predictions of subcellular location, transmembrane domains, signal peptide and GPI-anchor, GO, endoplasmic reticulum retention, and N-glycosylation domains. It also integrates results from InterProScan, PANTHER, Pfam, and alignment-based annotation searches. Additionally, the software provides a dataset of proteins with evidence of membrane localization, which is important for immunogenicity studies during vaccine development (Cheng et al., 2021; Kis et al., 2018) and diagnostic tests, such as ELISA (De Haro-Cruz et al., 2019; Iha et al., 2022) and western blotting (Begum, Murugesan & Tangutur, 2022; Crescitelli, Lässer & Lötvall, 2021; Springhorn & Hoppe, 2019).

FastProtein outputs files in formats that are widely used in the scientific community, including TSV, XLS, and ODF, as well as high-quality 300 DPI images, which is a widely used standard.

The total execution time of FastProtein depends on the InterProScan functional analysis and the local alignment method. By default, DIAMOND was selected due to its relatively faster execution time, although BLASTp is also available. The global median time for proteome-wide analyses was approximately 116 proteins per minute. This was increased to approximately 142 proteins per minute using DIAMOND, and approximately 90 proteins per minute. Thus, proteomic data from a large dataset can be quickly obtained using FastProtein. Considering differences in execution time and sensitivity, DIAMOND was chosen as the local aligner for the web server. DIAMOND can significantly decrease the alignment time (Buchfink, Reuter & Drost, 2021).

The only requirement for using the FastProtein software is the installation of Docker for local runs. FastProtein is an easy-to-use and viable tool for researchers with no background in bioinformatics because it provides a user-friendly interface similar to well-established software such as Blast2GO (Conesa et al., 2005), MEGA11 (Tamura, Stecher & Kumar, 2021), and MaxQuant (Prianichnikov et al., 2020). It also contributes to the initiatives that aim to democratize access to bioinformatics, such as the BioLib (https://biolib.com) and Galaxy (The Galaxy Community et al., 2024) projects.

Conclusions

FastProtein is a novel and user-friendly pipeline tool for proteomic data analysis that is available for small datasets and proteome-wide studies. Furthermore, it can be used through the CLI mode or a web interface. FastProtein accelerates proteomics analysis routines by generating multiple results in a one-step run. One of the limitations of FastProtein is that it does not yet integrate mass spectrometry data. However, the integration of both raw MS/MS data and data from other protein identification software through mass spectrometry is currently being implemented. The software is open-source and available at https://github.com/bioinformatics-ufsc/FastProtein, along with installation and execution instructions and test files generated for validation.

Supplemental Information