muddy_mine

Synopsis

Mud volcanoes represent geological structures, which host various hydrocarbonoclastic microbial consortia. Even though mud volcanoes can provide valuable data regarding hydrocarbon oxidation, topic is still off the radar. Meanwhile, NLP has been gaining traction over the last years in the mainstream fields (Wang et al. 2020). Niche environmental topics lamentably lag behind.

We present a mining pipeline - muddy_mine, which could engage NLP technology in the niche environmental topics such as mud volcanoes. This pipeline is able to mine taxonomy (bacterial, archaeal), methods or any other tokens of interest from the Open Access articles. Articles available in the S2ORC database, CC BY-NC 2.0, unmodified (Lo et al. 2020). muddy_mine output represents a csv table with all the relevant data regarding mud volcanoes.

muddy_mine was used to create muddy_db. muddy_db being the first biologically-oriented mud volcano database.

Check muddy_db web app: muddy_db

Check the published PeerJ article.

Cite this work Remizovschi A, Carpa R. 2021. Biologically-oriented mud volcano database: muddy_db. PeerJ 9:e12463 https://doi.org/10.7717/peerj.12463.

Methods

In order to aggregate biologically-oriented tokens, we used ScispaCy (Neumann et al. 2019) models. Taxonomy-flavored tokens were checked against NCBI Taxonomy database (Nov, 2020) (Schoch et al. 2020) . We built a local NCBI database with ETE3 (Huerta-Cepas, Serra, and Bork 2016).

Set up environment

• Download repository

git clone https://github.com/TracyRage/muddy_mine.git && cd muddy_mine

• Install miniconda and bioconda (optional)

• Initialize conda environment

conda env create --file environment.yml

• Activate conda environment

conda activate muddy_db

• Install ScispaCy model en_core_sci_sm

pip install https://s3-us-west-2.amazonaws.com/ai2-s2-scispacy/releases/v0.3.0/en_core_sci_sm-0.3.0.tar.gz

• Install NCBI Taxonomy database

python -c 'from ete3 import NCBITaxa; ncbi = NCBITaxa(); ncbi.update_taxonomy_database()'

If you get an error, click here.

Nota bene

S2ORC database includes around 12M papers. The full database has around 200GB (compressed). In order to avoid wrangling this heavy dataset, we created a subset of S2ORC data. You can find it in the sample_data/ directory. Pipeline demo works with that very subset.

If you need help, python any_muddy_script.py --help

Pipeline

1. Extract meta entries from the S2ORC (scan_s2orc_meta.py)

If you want to get articles from the S2ORC, then you need to have their metadata, be it PMIDs (Pubmed IDs) or DOIs. muddy_mine uses PMIDs to retrieve metadata from the S2ORC.

scan_s2orc_meta.py does the following procedures: (1) takes your Pubmed query (--entrez-query); (2) gets the correspondent PMIDs from the Pubmed; (3) decompresses S2ORC archives (--archives_path); (4) compares your list of PMIDs against S2ORC metadata; (5) writes the matching hits to an output file (--output_file).

2. Extract pdf entries from S2ORC

When you have metadata entries, you can proceed to extract the articles.

scan_s2orc_pdf.py does the following procedures: (1) uses the previous step output file (--metadata_input); (2) checks it against the S2ORC pdf_parse database (--pdf_archives); (3) writes the matching hits to an output file (--output_file).

3. Extract fields of interest from the jsonl files (extract_s2orc.py)

Outputs from the previous steps represent jsonl files. Those json files contain a lot of metadata fields, ranging from PMIDs to arxiv_ids. We need to extract the important ones.

extract_s2orc.py does the following procedures: (1) takes metadata (--metadata_input) and pdf_parse (--pdf_input) output files from the previous step and (2) creates files with essential fields (pmid, title, authors, abstract etc.) (--meta_output and --pdf_output).

4. Tabulate jsonl files (tabulate_s2orc.py)

When you have clean metadata and pdf_parse jsonl files, you can merge them and get a csv table. Use output files from the previous step (--meta_input, --pdf_input).

5. Mine data (mine_data.py)

When you have the csv table, you can mine all the relevant mud volcano data. Check module/get_terms.py and module/get_dict_terms.py for more information.

mine_data.py does the following procedures: (1) parses articles (--input_table); (2) mines and counts taxonomic-flavored tokens; (3) mines and counts chemical, geological and other tokens of interest; (3) writes results to csv tables (--output_mv and --output_taxa).

Nota bene

Mining process takes a lot of time. Check the mining results in the sample_data/mining_results/ directory.

References

Huerta-Cepas, Jaime, François Serra, and Peer Bork. 2016. “ETE 3: Reconstruction, Analysis, and Visualization of Phylogenomic Data.” Molecular Biology and Evolution 33 (6): 1635–38. https://doi.org/10.1093/molbev/msw046.

Lo, Kyle, Lucy Lu Wang, Mark Neumann, Rodney Kinney, and Daniel Weld. 2020. “S2ORC: The Semantic Scholar Open Research Corpus.” In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 4969–83. Online: Association for Computational Linguistics. https://doi.org/10.18653/v1/2020.acl-main.447.

Neumann, Mark, Daniel King, Iz Beltagy, and Waleed Ammar. 2019. “ScispaCy: Fast and Robust Models for Biomedical Natural Language Processing.” In Proceedings of the 18th BioNLP Workshop and Shared Task, 319–27. Florence, Italy: Association for Computational Linguistics. https://doi.org/10.18653/v1/W19-5034.

Schoch, C. L., S. Ciufo, M. Domrachev, C. L. Hotton, S. Kannan, R. Khovanskaya, D. Leipe, et al. 2020. “NCBI Taxonomy: a comprehensive update on curation, resources and tools.” Database (Oxford) 2020 (January).

Wang, J., H. Deng, B. Liu, A. Hu, J. Liang, L. Fan, X. Zheng, T. Wang, and J. Lei. 2020. “Systematic Evaluation of Research Progress on Natural Language Processing in Medicine Over the Past 20 Years: Bibliometric Study on PubMed.” J Med Internet Res 22 (1): e16816.