Serial KinderMiner (SKiM) is general literature-based discovery system for querying all ~32 million PubMed abstracts.
This repository provides the python script for SKiM, a general literature-based discovery system for uncovering unknown public knowledge from all ~32 million PubMed articles. This project is by the Stewart Computational Biology Group (https://morgridge.org/research/regenerative-biology/bioinformatics/) at the Morgridge Institute for Research, Madison, WI, USA.
A backend PubMed text index is required for executing SKiM. The code for building the backend PubMed text index is available at https://github.com/iross/km_indexer. In SKiM_chtc_query2.py, the user should assign their backend PubMed text index for the variable URL_BASE in line 13. The user should change the username and password in line 14, if applicable.
We applied SKiM for repurposing drugs for four diseases from Swanson's work, Raynaud's disease, migraine, Alzheimer's disease and schizophrenia. We compiled three lexicons from various resources for the study. The diseases lexicon was compiled from UMLS Metathesaurus and SNOMED CT (https://github.com/stewart-lab/Diseases_lexicon). The drugs lexicon was compiled from UMLS Metathesaurus, DrugBank and PharmGKB (https://github.com/CutaneousBioinf/LiteratureMiningTool/tree/master/DrugDict). For drug repurposing, we used a subset of drugs from the drugs lexicon (https://github.com/stewart-lab/Drugs_subset_for_drugs_repurposing_application). The phenotypes and synonyms lexicon was compiled from Human Phenotypes Ontology (HPO), Phenome Wide Association Studies (PheWAS), and Online Mendelian Inheritance in Man (OMIM) (https://github.com/stewart-lab/Phenotypes_and_symptoms_lexicon). Resources such as UMLS Metathesaurus requires license. Instead of sharing the lexicons, we share the scripts for compiling the lexicons.
We compiled a list of disease-drug associations from Comparative Toxicogenomics Database (CTD), National Drug File - Reference Terminology (NDF-RT), DrugBank and ClinicalKey (https://github.com/stewart-lab/Expert_curated_disease_drug_associations). We used our diseases and drugs lexicon to identify the duplicates across the resources. Instead of sharing the associations, we share the scripts for generating disease-drug associations from various resources.
The code for simulation study to find false discovery rate (FDR) is at https://github.com/zijianni/SKiM_simulation.
Display help:
$ python production_SKiM.py -h
Basic syntax for executing SKiM:
$ python production_SKiM.py keyphrase level_1_file level_2_file output_dir num_level2_queries
Example command using the sample A, B and C term files within 'data' directory: $ python production_SKiM.py data/A_term_file.txt data/B_terms_file.txt data/C_terms_file.txt sample_output/ 4
Output is saved within 'sample_output' directory. This directory contains L1_output, L2_output, and SKiM_output directories. The output from level 1 execution (A to Bs) are saved in L1_output directory. The output from level 2 execution (top n Bs to Cs, where n is number) are saved in L2_output directory. The final output is saved in SKiM_output directory. L1_output includes 'queries' directory to save various counts between A and every B, and 'with_ratios' directory to save significantly associated Bs with A. L2_output includes three directories: 'queries' directory to save various counts between every B from top n Bs and Cs, 'with_ratios' directory to save significantly associated Cs with every B, and 'all_bc_output' directory to save all B to C associations and significant Bs to Cs associations. L1_output, L2_output, and SKiM_output directories and their sub directories are created automatically during runtime. Please see sample_output directory for details.
SKiM was developed with Python 3.7.2.
Authors: Kalpana Raja and John Steill
Affiliation: Morgridge Institute for Research, Madison, WI, USA.