MINE Databases

The MINE Database contains code for generating (through Pickaxe) and storing and retrieving compounds from a database. Pickaxe applies reaction rules, representing reaction transformation patterns, to a list of user-specified compounds in order to predict reactions.

Documentation

For general information on MINE Databases, please consult JJeffryes et al. 2015..

Documentation is hosted at https://mine-database.readthedocs.io/en/latest/. It gives more detailed descriptions and example uses of the software.

Installation (current)

You need the uv package manager. If you don't have it, install it with the following command:

# On macOS and Linux.
curl -LsSf https://astral.sh/uv/install.sh | sh

# On Windows.
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Also be sure to install MongoDB (https://docs.mongodb.com/manual/installation/) if you are using the write_db = True option in Pickaxe. Installation information on Ubuntu system https://www.mongodb.com/docs/manual/tutorial/install-mongodb-on-ubuntu/

Running Pickaxe

Example Running Pickaxe through the CLI

Using MetaCyc rules and coreactants.

uv run mine_database/pickaxe_commons.py -C ./mine_database/data/metacyc_rules/metacyc_coreactants.tsv -r ./mine_database/data/metacyc_rules/metacyc_generalized_rules.tsv  -g 1 -c ./example_data/lotus_10.csv -o ./data/

Using original rules and coreactants (Enzymatic)

uv run mine_database/pickaxe_commons.py \
    --coreactant_list ./mine_database/data/original_rules/EnzymaticCoreactants.tsv \
    --rule_list ./mine_database/data/original_rules/EnzymaticReactionRules.tsv \
    --generations 1 \
    --compound_file ./example_data/Solanum_1000.csv \
    --output_dir ./data/ \
    --processes 8 \
    --explicit_h \
    --database testing-mine

or

uv run mine_database/pickaxe_commons.py -C ./mine_database/data/original_rules/EnzymaticCoreactants.tsv -r ./mine_database/data/original_rules/EnzymaticReactionRules.tsv  -g 1 -c ./example_data/lotus_1000.csv -o ./data/ -m 60

Complete command (writing to database):

uv run mine_database/pickaxe_commons.py --coreactant_list ./mine_database/data/original_rules/EnzymaticCoreactants.tsv --rule_list ./mine_database/data/original_rules/EnzymaticReactionRules.tsv --generations 1 --compound_file ./data/230106_lotus_inchikey_smiles.csv --output_dir ./data/ --processes 20  --explicit_h --database lotus_expanded --write_core

Command without writing to database:

uv run mine_database/pickaxe_commons.py --coreactant_list ./mine_database/data/original_rules/EnzymaticCoreactants.tsv --rule_list ./mine_database/data/original_rules/EnzymaticReactionRules.tsv --generations 1 --compound_file ./example_data/lotus_10.csv --output_dir ./data/ --processes 60 --verbose --explicit_h

Command with generalized rules and full LOTUS dataset without writing to database:

uv run mine_database/pickaxe_commons.py --coreactant_list ./mine_database/data/metacyc_rules/metacyc_coreactants.tsv --rule_list ./mine_database/data/metacyc_rules/metacyc_generalized_rules.tsv --generations 1 --compound_file ./data/LOTUS_inputfiles/230106_frozen_metadata_inchy_smiles.csv --output_dir ./data/ --processes 60 --verbose --explicit_h

Command loading target compounds and pruning the network to reactions that lead to them (--target_file needs an id column and a SMILES/InChI column; --prune_to_targets and --react_targets require --target_file). By default a generated compound that matches a target is not reacted further; add --react_targets to keep expanding it:

uv run mine_database/pickaxe_commons.py \
    --coreactant_list ./tests/data/test_coreactants.tsv \
    --rule_list ./tests/data/test_single_reaction_rule.tsv \
    --generations 2 \
    --smiles 'Cc1cc2c(cc1C)N(CC(O)C(O)C(O)COP(=O)(O)OP(=O)(O)OCC1OC(n3cnc4c(N)ncnc43)C(O)C1O)c1nc(O)nc(O)c1N2' \
    --target_file ./tests/data/test_targets.csv \
    --prune_to_targets \
    --output_dir ./data/ \
    --explicit_h

--rule_list can be repeated to load several rule files together. mine_database/data/metacyc_rules/metacyc_custom_rules.tsv holds hand-written rules that are not in the published generalized set: chalcone synthase (rule_CHS: acyl-CoA + 3 malonyl-CoA → chalcone + 4 CoA + 3 CO2) and flavonol synthase (rule_FLS: dihydroflavonol + 2-oxoglutarate + O2 → flavonol + succinate + CO2 + H2O). They need metacyc_coreactants.tsv (which defines MALONYL-COA and SUCCINATE) and run without --explicit_h:

uv run mine_database/pickaxe_commons.py \
    --coreactant_list ./mine_database/data/metacyc_rules/metacyc_coreactants.tsv \
    --rule_list ./mine_database/data/metacyc_rules/metacyc_generalized_rules.tsv \
    --rule_list ./mine_database/data/metacyc_rules/metacyc_custom_rules.tsv \
    --generations 1 \
    --compound_file ./example_data/start_building_blocks.tsv \
    --output_dir ./data/

Filtering and pruning between generations

Filters run on every generation before it is expanded; compounds they reject are not expanded and are removed from the network when nothing else depends on them. --prune_between_gens additionally prunes the network after filtering to what leads to compounds that are still expanded (and targets).

Option	Filter	Notes
--similarity_cutoff 0.3 or 0,0.26,0.29	Tanimoto (RDKit fingerprint) to the targets	One value, or one per generation (last value reused). Needs --target_file.
--similarity_sample N	Keep N compounds per generation, sampled weighted by similarity	Runs after all cutoff filters and samples only what they kept. Compounds with similarity ≤ 0.15 are never sampled. Needs --target_file.
--mcs_cutoff 0.5	Maximum common substructure with the targets (fraction of atoms + bonds)	One value or one per generation; slow on large molecules. Needs --target_file.
--min_mw, --max_mw	Molecular weight range
--atomic_composition 'C:5-30,N:0-0'	Atom-count ranges	Empty bound is open (C:5-).

Generation 0 (the starting compounds) is filtered too, so give per-generation cutoffs starting at 0 if early intermediates are dissimilar to the target. Example (naringenin chalcone → quercetin):

uv run mine_database/pickaxe_commons.py \
    --coreactant_list ./mine_database/data/metacyc_rules/metacyc_coreactants.tsv \
    --rule_list ./mine_database/data/metacyc_rules/metacyc_generalized_rules.tsv \
    --rule_list ./mine_database/data/metacyc_rules/metacyc_custom_rules.tsv \
    --smiles 'Oc1ccc(/C=C/C(=O)c2c(O)cc(O)cc2O)cc1' \
    --generations 4 --processes 8 \
    --target_file ./example_data/target_quercetin.tsv \
    --similarity_cutoff 0,0.26,0.29,0.30 --similarity_sample 200 \
    --prune_between_gens --prune_to_targets \
    --output_dir ./data/

Accessing the mongodb database

To access the mongodb database, you can use the following command:

mongosh

Then, you can use the following commands to access the database:

use example_db
db.compounds.find()

Optionally the DB can be observed using MongoDB Compass. See https://www.mongodb.com/docs/compass/current/install/