Skip to content
This repository has been archived by the owner on Jul 3, 2023. It is now read-only.

MolecularAI/ReinventCommunity

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

54 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ReinventCommunity (jupyter notebook tutorials for REINVENT 3.2)

This repository is a collection of useful jupyter notebooks, code snippets and example JSON files illustrating the use of Reinvent 3.2. At the moment, the following notebooks are supported:

  • Complete_Use-Case_DRD2_Demo: a full-fledged use case using public data on DRD2, including use of predictive models and elucidating general considerations
  • Create_Model_Demo: explanation on how to initialize a new model (prior / agent) for REINVENT which can be trained in a transfer learning setup
  • Data_Preparation: tutorial on how to prepare (clean, filter and standardize) data from a source such as ChEMBL to be used for training
  • Model_Building_Demo: shows how to train a predictive (QSAR) model to be used with REINVENT based on the public DRD2 dataset (classification problem)
  • Reinforcement_Learning_Demo: example reinforcement learning run with a selection of scoring function components to generate novel compounds with ever higher scores iteratively
  • Reinforcement_Learning_Demo_Selectivity: example illustrating the use of the relatively complicated selectivity_component to optimize potency against a target while simultaneously pushing for a low potency against one or more off-targets
  • Reinforcement_Learning_Demo_Tanimoto: very simple (only 1, easy-to-understand component) transfer learning example
  • Reinforcement_Learning_Exploitation_Demo: illustrates the exploitation scenario, where one is after solutions from a subspace in chemical space already well defined
  • Reinforcement_Learning_Exploration_Demo: illustrates the exploration scenario, where the aim is to generate a varied set of solutions to a less stringently defined problem
  • Reinforcement_Learning_Demo_DockStream: illustrates the use of DockStream in REINVENT, allowing the generative model to gradually optimize the docking score of proposed compounds. For more information on DockStream, please see the DockStream repository and the corresponding DockStreamCommunity repository for tutorial notebooks on DockStream as a standalone molecular docking tool.
  • Reinforcement_Learning_Demo_Icolos: illustrates the use of Icolos in REINVENT using a docking scenario.
  • Sampling_Demo: once an agent has been trained and is producing interesting results, it can be used to generate more compounds without actually changing it further - this is facilitated by the sampling mode
  • Score_Transformations: as many components produce scores on an arbitrary scale, but REINVENT needs to receive it normalized to be a number between 0 and 1 (with values close to 1 meaning "good"), score transformations have been implemented and can be used as shown in this tutorial
  • Scoring_Demo: in case a set of existing compound definitions (for example prior to starting a project) should be scored with a scoring function definition, the scoring mode can be used
  • Transfer_Learning_Demo: this tutorial illustrates the transfer learning mode, which usually is used to "pre-train" an agent before reinforcement learning in case no adequate naive prior is available or to focus an already existing agent further
  • Transfer_Learning_Demo_Teachers_Forcing: same as Transfer_Learning_Demo above, with explanation of teachers forcing
  • Lib-INVENT_RL1_QSAR: Lib-INVENT example reinforcement learning run using a QSAR model
  • Lib-INVENT_RL2_QSAR_RF: Lib-INVENT example reinforcement learning run using a random forest (RF) QSAR model
  • Lib-INVENT_RL3_ROCS_RF: Lib-INVENT example reinforcement learning using OpenEye's ROCS 3D similarity (requires an OpenEye license)
  • Link-INVENT_RL: Link-INVENT example reinforcement learning
  • Automated_Curriculum_Learning_demo: illustrates the automated curriculum learning running model. The example demonstrates how to set-up a curriculum to guide the REINVENT agent to sample a target molecular scaffold. This scenario represents a complex objective as the target scaffold is not present in the training set for the prior model