package rankers

  1. Overview
  2. Docs
Vanishing Ranking Kernels (VRK)

Install

Dune Dependency

Authors

Maintainers

Sources

v1.0.0.tar.gz
sha256=623ae59cc4a04703f2424464241afc4fc542e86abd51d4a101e93a1276d41174
md5=43b6ac3433d6ff75302a3576b6ae067a

Description

Reference implementation of the Vanishing Ranking Kernels method.

A single parameter QSAR modeling technique for HTS data; with an applicability domain.

Manuscript to appear soon.

Published: 20 Nov 2019

README

RanKers

Reference implementation of the Vanishing Ranking Kernels (VRK) method

Example

Example ROC curve on a hold-out test set. The test set had 38 active molecules and 664 inactives. ROC AUC: 0.861; BEDROC AUC: 0.766; PR AUC: 0.678. The ROC curve is in purple; the precision-recall (PR) curve in cyan. The probability of activity given a raw score is the red curve. The green curve is the number of actives divided by the number of decoys as a function of the scores filtering threshold.

Train and test a model:

rankers_bwmine -i data/tox21_nrar_ligands_std_rand_01.txt

Same, but using 16 cores :

rankers_bwmine -np 16 -i data/tox21_nrar_ligands_std_rand_01.txt

Usage

rankers_bwmine -i <train.txt>
  [-p <float>]: proportion of the (randomized) dataset
  used to train (default=0.80)
  [-k {uni|tri|epa|biw}]: kernel function choice (default=biw)
  [-np <int>]: max number of processes (default=1)
  [-o <filename>]: write raw test scores to file
  [--train <train.txt>]: training set (overrides -p)
  [--valid <valid.txt>]: validation set (overrides -p)
  [--test <test.txt>]: test set (overrides -p)
  [-n <int>]: max number of optimization steps; default=150
  [--capf <float>]: keep only fraction of decoys
  [--capx <int>]: keep only X decoys per active
  [--capi <int>]: limit total number of molecules
  (but keep all actives)
  [--seed <int>: fix random seed]
  [--pr]: use PR AUC instead of ROC AUC during optimization
  [-kb <float>]: user-chosen kernel bandwidth
  [--mcc-scan]: scan classif. threshold to maximize MCC
  [--tap]: tap the train-valid-test partitions to disk
  [-q|--quick]: exit early; just after model training
  [--noplot]: turn off gnuplot
  [-v]: verbose/debug mode
  [-h|--help]: show this help message

Dependencies (12)

  1. parmap
  2. parany >= "6.0.0" & < "10.0.0"
  3. nlopt-ocaml
  4. molenc
  5. minicli
  6. dune >= "1.6"
  7. dolog >= "4.0.0" & < "5.0.0"
  8. cpm >= "10.0.0"
  9. conf-gnuplot
  10. bst
  11. batteries
  12. base-unix

Dev Dependencies

None

Used by

None

Conflicts

None

OCaml

Innovation. Community. Security.