FuncFetch

The development of the FuncFetch workflow was funded by the National Science Foundation IOS-Plant Genome Research Program Award #2310395. Additional project details can be found on the Moghe Lab Tools website. This Repository is meant to version-control and share the scripts and the workflow. The outputs of this workflow can be found on the Moghe Lab Tools website as well as the FuncZymeDB repository

FuncFetch workflow scripts

Step 1: Retrieve paper titles and abstracts from PubMed

Edit the Step 1 config file. You can find the template config file in the config folder in this repository.

[entrez]
email = ENTER_YOUR_EMAIL
api_key = YOUR_PUBMED_API_KEY
requests_per_second = 200
tool = biopython

[query_settings]
query = YOUR_ENZYME_FAMILY_QUERY

[journal_list]
journallist = main_journalsList.tab
reviewfilter = yes

[keywords_list]
keywords_list = main_keywordsList.tab

[use_elink]
use_elink = no

Run the script

python funcfetch_step1.py --config bahd_step1_efetch.config

This will retrieve the following files to your working directory

BAHD_acyltransferase_OR_hydroxycinnamoyl-step1-ncbiAbstracts.txt
BAHD_acyltransferase_OR_hydroxycinnamoyl-step1-ncbiQuery.txt
BAHD_acyltransferase_OR_hydroxycinnamoyl-step1-ncbiSummary-failedScreen.txt
BAHD_acyltransferase_OR_hydroxycinnamoyl-step1-ncbiSummary.txt

The .log file will note the parameters used and the output

Your selections are as follows:
CONFIG_FILE: bahd_step1_efetch.config
EMAIL: ....
APIKEY: ....
TOOL: biopython
QUERY: BAHD acyltransferase OR hydroxycinnamoyl
JOURNALS_LIST: main_journalList.tab
FILTER REVIEWS: yes
KEYWORDS_LIST: main_keywordsList.tab.fil
USE_ELINK: no

Output stats:
# of hits to Initial query: 667
Total # of hits to Initial query: 667
# of abstracts in Initial after all filters applied: 461

Some of these files will be passed on to Step 2

Step 2: Use LLM to screen abstracts

If you don't already have an account with OpenAI, make one. https://platform.openai.com/signup

Navigate to the OpenAI API platform dashboard and generate a secret key under "API Keys" if you don't already have one. Per OpenAI's recommendations: "Do not share your API key with others, or expose it in the browser or other client-side code." This API key will be required to run Step 2 and Step 4, so record it in a secure location.

Edit the Step 2 config file. You can find the template config file in the config folder in this repository.

[openai]
key = YOUR_OPENAI_API_KEY
# The organization field is optional and only necessary if your openai user account belongs to multiple organizations.
organization = 

[rate_limit]
# Request limits may vary depending on the OpenAI account and policies. For more information on limits, look under "Organization" "Limits" in the settings section of the OpenAI API platform.
requests_per_minute = 5000

[model_settings]
# We tested the FuncFetch workflow with the gpt-4-turbo-2024-04-09 model, but feel free to swap in a more recent model.
model = gpt-4-turbo-2024-04-09
temperature = 0.5

[query_settings]
# We recommend using the same query for all workflow steps, as this replicates the testing conditions.
query = YOUR_ENZYME_FAMILY_QUERY

[questions]
question_11 = Is this likely a paper that describes the biochemical (i.e. catalytic) activity of a {query} enzyme? Ideally, this confirmation would include an in vitro biochemical assay. Answer only with a 1 or 0, corresponding to yes or no respectively.
question_10 = Is this likely a paper that demonstrates the biochemical (i.e. catalytic) activity of a {query} enzyme? Ideally, this confirmation would include an in vitro biochemical assay. Answer only with a 1 or 0, corresponding to yes or no respectively.
question_9 = Is this likely a paper that directly tests the biochemical (i.e. catalytic) activity of a {query} enzyme? Ideally, this confirmation would include an in vitro biochemical assay. Answer only with a 1 or 0, corresponding to yes or no respectively.
question_6 = Is this abstract likely to describe a paper that directly tests the biochemical (i.e. catalytic) activity of a {query} enzyme? Ideally, this confirmation would include an in vitro biochemical assay. Answer only with a 1 or 0, corresponding to yes or no respectively.
question_3 = Is this abstract likely to describe a paper that conducted a direct assay of a {query} enzyme catalyzing a reaction with a chemical substrate? Answer only with a 1 or 0, corresponding to yes or no respectively.

Run the script

python funcfetch_step2.py --config funcfetch_step2.config --abstracts ENZYME_FAMILY-step1-ncbiAbstracts.txt --summary ENZYME_FAMILY-step1-ncbiSummary.txt

# For details on arguments and parameters run this line:
python funcfetch_step2.py -h

This step will produce some JSONL files that are the input and output for the OpenAI batch API run. They will look like the following:

• ENZYME_FAMILY_batch_input.jsonl
• ENZYME_FAMILY_batch_output.jsonl

Additionally, the script with produce human and machine readable files detailing the filtered papers and their score as relevant or irrelevant:

• ENZYME_FAMILY_filtered_abstracts.tsv
• ENZYME_FAMILY_filtered_abstracts.txt

The RIS format file is the most important output and required as input into Zotero for Step 3:

• ENZYME_FAMILY_relevant_papers.ris

Step 3: Get papers using Zotero

• Create a new collection in Zotero
• Load the RIS file generated by Step 2 into the collection
• Once the library is created, select all articles, right click and select "Find PDFs..."
• Most of the papers will be downloaded automatically
• For the remaining papers, click on the DOIs, go to the page, download the PDF, drag the PDF into the Zotero entry manually.
• Once all PDFs are obtained, export the collection as a csv file (the default)
• Navigate to the advanced settings section of Zotero
• Note the "Data Directory Location" under the Files and Folders tab

Step 4: Use LLM to extract activity and metadata

Edit the Step 4 config file. You can find the template config file in the config folder in this repository.

[openai]
key = YOUR_OPENAI_API_KEY
# The organization field is optional and only necessary if your openai user account belongs to multiple organizations.
organization = 

[rate_limit]
# Request limits may vary depending on the OpenAI account and policies. For more information on limits, look under "Organization" "Limits" in the settings section of the OpenAI API platform.
requests_per_minute = 500

[model_settings]
# We tested the FuncFetch workflow with the gpt-4o-2024-05-13 model, but feel free to swap in a more recent model.
model = gpt-4o-2024-05-13
temperature = 0.5

[query_settings]
# We recommend using the same query for all workflow steps, as this replicates the testing conditions.
query = YOUR_ENZYME_FAMILY_QUERY

[step4_instructions]
text_pdf = Objective: Precisely extract and present biochemical data on enzyme family members from {form_ref_0}, using a specified JSON structure ...
merge = Objective: Merge biochemical data from two distinct JSON object lists ...

Run the script

python funcfetch_step4.py --config funcfetch_step4.config --keys YOUR_EXPORTED_ZOTERO_COLLECTION --zotero_storage YOUR_DATA_DIRECTORY_LOCATION --current_storage PATH_TO_DATA_DIRECTORY_LOCATION

# For details on arguments and parameters run this line:
python funcfetch_step4.py -h

This step will produce some log files with all messages recieved from the Open API for text and pdf method steps, and any entries recieved that don't pass validation of all fields:

• messages.log
• invalid_entries.log

The final products of Step 4 are three tabular files of entries generated by each method:

• ENZYME_FAMILY_merge_method.tsv
• ENZYME_FAMILY_pdf_method.tsv
• ENZYME_FAMILY_text_method.tsv

(Optional) Flag Step 4 Entries with Potential Issues

We have included a helper script that can identify output entries where one of the text or pdf methods, or both methods, produce no output. This can potentially help identify cases of hallucinations or other errors, but flagged entries are not necessarily erroneous. This script adds an additional column to the tabular file with one of the following for each entry: "check1", "check2", "pass", and "pass_noDOI". The "check" entries are flagged entries, and the original API messages are written to files with the extension "_part_#.txt". Entries with "pass", passed the flagging, and "pass_noDOI" entries are those with no DOI so pass with no flagging by default.

Run the script

python flag_funcfetch_step4_output.py messages.log ENZYME_FAMILY_merge_method.tsv

Step 5: Filter

python funcfetch_step5.py species.taxonomy BAHD_acyltransferase_or_hydroxycinnamoyl_transferase_merge_method.flag.tsv
Reading Species info...
Parsing FuncFetch output file...
>>>Missing: Saccharothrix_espanensis --> Taxonomy added as NA
>>>Missing: Saccharothrix_espanensis --> Taxonomy added as NA
>>>Missing: Dendranthema_× --> Taxonomy added as NA
>>>Missing: Dendranthema_× --> Taxonomy added as NA
>>>Missing: Apple_tree --> Taxonomy added as NA
>>>Missing: Eggplant_(Solanum --> Taxonomy added as NA
>>>Missing: Chicory_(Cichorium --> Taxonomy added as NA
>>>Missing: Dendranthema_× --> Taxonomy added as NA
>>>Missing: Dendranthema_× --> Taxonomy added as NA
>>>Missing: Dendranthema_× --> Taxonomy added as NA
>>>Missing: Dendranthema_× --> Taxonomy added as NA
>>>Missing: Dendranthema_× --> Taxonomy added as NA
>>>Missing: Physcomitrella_patens --> Taxonomy added as NA
>>>Missing: Lycopersicon_esculentum --> Taxonomy added as NA
>>>Missing: Lycopersicon_esculentum --> Taxonomy added as NA
>>>Missing: Lycopersicon_esculentum --> Taxonomy added as NA
>>>Missing: Lycopersicon_esculentum --> Taxonomy added as NA
>>>Missing: Lycopersicon_esculentum --> Taxonomy added as NA
>>>Missing: Lycopersicon_esculentum --> Taxonomy added as NA
>>>Missing: Lycopersicon_esculentum --> Taxonomy added as NA
>>>Missing: Lycopersicon_esculentum --> Taxonomy added as NA
>>>Missing: Physcomitrella_patens --> Taxonomy added as NA
>>>Missing: Physcomitrella_patens --> Taxonomy added as NA
>>>Missing: Physcomitrella_patens --> Taxonomy added as NA
# of species with activity info: 1527
# of plant species with activity info: 1455
# of bacterial species with activity info: 15
# of other species with activity info: 57
Done!

This script will create two files -- a .log file and a .tax file. The .tax file will have the taxonomic associations in a tabular format

TITLE   DOI     SPECIES Family  Kingdom ENZYME_COMMON_NAME      ENZYME_FULL_NAME        GENBANK UNIPROT_ID      ALT_ID  SUBSTRATE       PRODUCT FLAG
Analysing a Group of Homologous BAHD Enzymes Provides Insights into the Evolutionary Transition of Rosmarinic Acid Synthases from Hydroxycinnamoyl-CoA:Shikimate/Quinate Hydroxycinnamoyl Transferases.    10.3390/plants13040512  Mentha longifolia       4136|Lamiaceae  33090|Viridiplantae     MlAT1   Hydroxycinnamoyl-CoA:Shikimate/Quinate Hydroxycinnamoyl Transferase     NA      NA      NA         p-coumaroyl-CoA; shikimate      p-coumaroyl shikimate   pass

This file can be opened in Excel to manually curate the rest of the way. This curation can be done in a stepwise manner. You may see the Excel files deposited in the Minimally Curated Sets folder to see how we performed this curation.

Questions/concerns

If you have any questions or concerns, or just want to thank us, email Nathaniel Smith (nss97) or Gaurav Moghe (gdm67). We have Cornell email addresses.