Skip to content

Latest commit

 

History

History
77 lines (63 loc) · 2.51 KB

BENCHMARK_CONFIG.md

File metadata and controls

77 lines (63 loc) · 2.51 KB
Raw

BENCHMARK CONFIGURATION FILE

File that both describes and specifies the details of the benchamrk run Simply edit the code-blocks (surrouned with ``` characters) in order to configure your specific parameters of the benchmark.

Datasets

Options for datasets are datasets I

DATASETS = [I]

The details of the dataset are below:

Dataset I

Dataset I was originally described in

S. Riniker, G. Landrum, J. Cheminf., 5, 26 (2013), DOI: 10.1186/1758-2946-5-26, URL: http://www.jcheminf.com/content/5/1/26

The dataset consists of 88 targets from MUV, DUD, and ChEMBL.

Fingerprints

Options for fingerprints are any subset of:

  • ecfp0, ecfp2, ecfp4, ecfp6 (1024 bit RDKit Morgan Fingerprint with diameter 0/2/4/6)
  • ecfc0, ecfc2, ecfc4, ecfc6 (unhashed RDKit Morgan Fingerprint with diameter 0/2/4/6)
  • fcfp2, fcfp4, fcfp6 (1024 bit feature-based RDKit Morgan Fingerprint with diameter 2/4/6)
  • fcfc2, fcfc4, fcfc6 (unhashed feature-based RDKit Morgan Fingerprint with diameter 2/4/6)
  • lecfp4, lecfp6 (16384 bit RDKit Morgan Fingerprint with diameter 4/6)
  • lfcfp4, lfcfp6 (16384 bit feature-based RDKit Morgan Fingerprint with diameter 4/6)
  • maccs (MACCS Keys ExplicitBitVect fingerprint)
  • ap (Atom Pairs IntSparseIntVect fingerprint)
  • hashap (1024 bit hashed Atom Pair fingerprint)
  • tt (Topological Torsions LongIntSparseIntVect fingerprint)
  • hashtt (1024 bit hashed Topological Torsion fingerprint))
  • avalon (1024 bit Avalon Fingerprint)
  • laval (16384 bit Avalon Fingerprint)
  • rdk5, rdk6, rdk7 (RDKit path-based fingerprint with path-legnth 5/6/7)
  • cas (CAS 7851 bit fingerprint. Note that cas data files are needed in /data to run cas)
FINGERPRINTS = [rdk6, cas]

Methods

Both classical machine learning (ML) methods and basic similarity metrics may be used for activity prediction using the provided fingerprints.

ML Methods

The provided ML algorithm options are any subset of

  • Logistic Regression (LR)
  • Naive Bayes (NB)
  • Random Forests (RF):
ML_methods = [LR, NB, RF]

The provided similarity metric options are any subset of

  • Tanimoto
  • Dice
  • Cosine
  • Russel
  • Kulczynski
  • McConnaughey
  • Manhattan
  • RogotGoldberg
SIMILARITY_methods = [Tanimoto]

Number of Query Molecules

In each assay, randomly selected subsets of the active and inactive compounds are used for "training" (note that no parameters are fit if using a similarity metric).

One decision is the number of query active molecules to use for training. Options are 5,10, or 20 active compounds.

NUM_QUERY_MOLS = 10