Introduction

AdaptiveFlow is a versatile, parallel workflow platform for carrying out virtual screening-related tasks on Linux-based computer clusters of any type and size which are managed by a batch system (such as SLURM). 

Currently, there exist two versions of AdaptiveFlow, which are tailored to different types of tasks:

AFLP and AFVS use the same core technology regarding the workflow management and parallelization, and they can be used individually or in concert with each other. Additional versions are expected to arrive in the future. 

Pre-built, ready-to-dock ligand libraries for AFVS are available for free (in the download section).

General Features

Heavenly Fast
AdaptiveFlow can be extremely fast due to its perfect scaling behavior (even when using very large number of CPUs). There are virtually no bounds regarding the number of processors which can be utilized by AdaptiveFlow. In addition, it supports some of the fastest docking programs available such as QuickVina 2.
Error Robustness
AdaptiveFlow is relatively robust regarding unexpected errors and interruptions, which often occur on computer clusters. AdaptiveFlow can respond to signals sent to it by the (batch or operating) system in the case of cluster problems, but even after termination without warning it can simply be resumed.
Any Cluster Configuration
AdaptiveFlow runs out of the box on almost any GNU/Linux cluster which is managed by a batch system. The workflow tool can use any conceivable hardware configuration regarding the number of cores/CPUs, the number of sockets, the number of nodes, etc. The software runs on any Linux distribution.
Cloud Ready
Cloud Compatible
Due to it's architecture, AdaptiveFlow is able to run on cloud computing platforms such as the Google Cloud Platform (GCP), Amazon Web Services (AWS), or Microsofts Azure. It works both with on-demand as well as the much cheaper preemptible virtual machines.
Multiple Batch Systems
AdaptiveFlow currently supports SLURM, the most widely used batch system on computer clusters and supercomputers. In addition, it supports AWS Batch for native AWS Cloud support.
Monitoring
The workflow can be monitored in real time during the execution. This allows tracking the progress of the workflow, as well as examining possible problems during or after the runtime.
Highly Automatized
AdaptiveFlow can run fully automatically for any duration of time until it has processed all ligands specified in the input files. It achieves this by autonomously ending and submitting new jobs into the batch system as needed.
Realtime Control
The workflow can be controlled and modified during the runtime. The workflow can, for instance, be paused, resumed, and the utilized hardware resources and job configurations changed. The workflow can be transferred to other clusters and resumed there if desired.
Convenient Input and Output Formats
The input and output ligand databases consist of hierarchical multilevel tar-archives which are compressed. This compact format allows to easily handle vast amounts of ligands in an efficient and scalable manner.
Free Software
AdaptiveFlow is free/libre software and licensed under the GNU GPL v2, which guarantees users certain rights (of freedom). Moreover, AdaptiveFlow is available at no cost. (Free/Libre software and Open Source are two distinct concepts.)
Open Source Development Model
AdaptiveFlow is developed in an open collaborative development model, warmly inviting anyone to join. This allows AdaptiveFlow to grow more quickly and healthily in the way the community desires.

 

AFVS - AdaptiveFlow for Virtual Screening

AFVS is dedicated to carrying out virtual screening procedures on computer clusters.

It can be applied in a number of different settings:

  • Hit identification by screening ligand libraries of any size
  • Hit optimization by screening custom hit-based analog libraries
  • Thorough and extensive dockings of one or more molecules

AFVS can be used seamlessly with AVLP when the ligand library needs to be prepared.  

AFVS Features

Choice of Docking Program
AFVS supports a large number of different docking programs. Sampling algorithms for various docking programs can be combined with different scoring functions from other docking programs, leading to 1500 different docking methods.
Multiple Replicas
Each docking scenario can be carried out multiple times. This can be very useful when one wants to increase the chance that the docking program finds the docking pose with the global minimum relative to the scoring function which it employs.
Multistaging
AFVS is suitable for carrying out virtual screenings in a multi-stage manner. In each stage the accuracy of the dockings can be increased to rescore the highest scoring compounds from the previous stage. This enhances the quality of the virtual screening at reduced computational costs. 
Multiple Docking Scenarios
AFVS allows for the carrying out of multiple docking scenarios per ligand. A docking scenario in AFVS is defined by the receptor structure, by the docking parameters (such as exhaustiveness), rigid or flexible receptor docking, the choice of flexible receptor side chains, or the docking program. This also allows for ensemble dockings.

 

AFLP - AdaptiveFlow for Ligand Preparation

AFLP is dedicated to the preparation of ligand databases in a ready-to-dock format. Moreover, it provides them in a format which is directly usable by AFVS. But AFLP can also be used independently of AFVS (and virtual screenings in general).

AFLP can use ChemAxon's JChemSuite for carrying out several steps of the molecule preparations, though Open Babel can currently be used as well instead. A suitable academic or commercial license needs to be obtained from ChemAxon directly. More information can also be found here in the documentation.

AFLP Features

Any Target Format
Almost every chemical file format (i.e.all formats supported by Open Babel) can be chosen for the final output files of the ligands. Among these file types are all the well known formats such as SDF or PDB, but also specialized ones such as PDBQT which is used by most AutoDock-based docking programs.
Ligand Protonation at any pH
During the preparation of the ligands with AFLP, they can be protonated at a freely choosable pH value. Ligand protonation is an optional step in the workflow, which does not have to be employed. It can be carried out either by ChemAxons cxcalc or by Open Babel.
3D Conformation Generation
Docking programs normally require three-dimensional conformations of the molecules. AFLP can prepare them when required, either by using OpenBabel or by using ChemAxon's molconvert.
Fallback Modes
Sometimes, one molecular conversion programs like OpenBabel or the tools from ChemAxon fail for certain molecules during one of the processing steps such as protonation. In AFLP one can specify a backup program in the case that the primary conversion program should fail during the conversion of some ligands.
Desalting and Neutralization
Molecules which are provided in salt-form can be automatically desalted and neutralized via ChemAxons JChem suite.
Tautomerization
Often multiple tautomeric states of the same compounds exist at a given pH. AFLP can automatically generate the tautomers using the cxcalc tool of ChemAxons JChem Suite, with the possibility to specify all options which the tautomer plugin of cxcalc provides. This can lead to a substantial increase in the size of the library.
Heavenly Fast
Substantial additional speedup due to the use of Nailgun, which runs a persistent JVM during the runtime of the workflow.
Multiple Output Formats
Any number of distinct output formats can be generated during a single workflow. For each of these formats, a separate output ligand library is created.

 

AFU - AdaptiveFlow

AdaptiveFlow Unity (AFU) combines ligand preparation and virtual screens in a unified workflow. 

AFU Features