chaff-tools 0.0.1

Pipeline for contamination, calcuating raco-score, and assessing sensitivty of both physic-based and machine-learning models.

What is it?

TBD

Installation

Installation made easy. chaff-tools is pip installable. Please make sure you have Python 3.6 installed.

$ pip install chaff-tools

Usage

Miscallenous tools are used

# Extract a ZIP file and flatten contents
$ chaff-physics extract-zip

# Extract a .db2.gz file
$ chaff-physics extract-db2

# Extract all .db2.gz files from a folder
$ chaff-physics extract-db2-folder

# Download and process TLDR batch jobs from a CSV
$ extract-db2-folder

# Extract all .db2.gz files from a folder
$ extract-db2-folder

Random split

The cp-random-split tool allows you to perform a realistic, clustering-based train/test split on a dataset of molecular SMILES, as described in the Martin et al. paper. This helps preserve chemical diversity between sets, unlike purely random splits.

Command

cp-random-split \
  --input-file compounds.csv \
  --smiles-col 0 \
  --frac-train 0.8 \
  --exact \
  --method Auto \
  --output results/some_split

Flag	Description
-i, --input-file	Path to the input CSV/TSV file containing SMILES.
-s, --smiles-col	Index of the column containing SMILES (default: 0).
-f, --frac-train	Fraction of the dataset to use for training (e.g. 0.8).
--exact	If set, may split a cluster to achieve the exact fraction.
-m, --method	Clustering method: Auto, TB (Butina), or Hierarchy (default: Auto).
-o, --output	Base path for output files; _train.csv and _test.csv will be created.

Co-existence with tldr-tools

tldr-tools can be synergistically used with chaff-tools to run larger-scale pipelines.

tldr-batch

$ chaff-contaminate --actives_dir path/to/actives --contaminants_dir path/to/contaminants --frac_contaminate 0.2 --output path/to/output.yaml --seed 42

Arguments:

--actives_dir: Directory containing .db2 files for actives.
--contaminants_dir: Directory containing .db2 files for contaminants.
--frac_contaminate: Fraction of actives that should be contaminated (value between 0 and 1).
--output: Path where the YAML file with results will be saved.
--seed: (Optional) Random seed for reproducibility.

For example, if running decoy generation is desired:

tldr-submit --module decoys --activesism input_files/actives.ism --decoygenin input_files/decoy_generation.in --memo "Decoy generation for ADA, replicate 1"

Or, you can build a ligand using DOCK3.8:

tldr-submit --module build --input chaff_tools/aggregator_advisor_hf_test.txt --memo "aa_hf_test"

Documenting runs with the optional memo parameter is encouraged.

Pass in a job number to check on a status of a run:

tldr-status --job-number 14886

Once a run is successful, you can download the output to a local directory:

tldr-download --job-number 14886 --output some_folder

Does tldr-tools work in Colab and Jupyter Notebook?

Yep, you use chaff-tools as follow:

TBD