Understanding Role of Provenance in Bioinformatics Workflows and Enabling Interoperable Computational Analysis Sharing

PhD Completion Seminar

Presenter: Farah Zaib Khan

Supervisors: AProf. Andrew Lonie, Prof. Richard O. Sinnott

Chair: Prof. Adrian Pearce

@farahzk03

https://slides.com/farahzkhan/phd-seminar

November 28th, 2018

Introduction - Context

  • Workflows

  • Role of Workflows in Bioinformatics

  • Why use workflows?

  • Provenance

  • Why should we care about Provenance?

  • Workflow Management Systems

@farahzk03

https://slides.com/farahzkhan/phd-seminar

Workflows

(Esp. scientific workflows)

"The description of a process for accomplishing a scientific objective, usually expressed in terms of tasks and their dependencies"

(Ludäscher et al. 2009)

 

  • Automated mechanism to systematise a computational analysis and capture the methods
  • Directed Acyclic Graphs
  • Data-flow oriented

 

@farahzk03

https://slides.com/farahzkhan/phd-seminar

Workflow Life Cycle

4 Stages

  • Composition, Representation and Data Model

  • Mapping to Resources

  • Workflow Execution

  • Metadata and Provenance

Introduction - Context

  • Workflows

  • Role of Workflows in Bioinformatics

  • Why use workflow?

  • Provenance

  • Why should we care about Provenance?

  • Workflow Management Systems  

@farahzk03

https://slides.com/farahzkhan/phd-seminar

Exponential advances in the technologies and  instruments

Declining sequencing cost ( From 2.7 billion USD to 1000USD)

  • Ad hoc methods (scripts, manual steps etc.) are not enough to handle the volume of data.

  • Computational Bioinformatics Workflows are employed to automate and record the steps performed.

Genomic data produced exponentially (tenfolds/year since 2002)

Now called four-headed beast...

Acquisition, Storage, Distribution, Analysis

@farahzk03

https://slides.com/farahzkhan/phd-seminar

Introduction - Context

  • Workflows

  • Role of Workflows in Bioinformatics

  • Why use workflow?

  • Provenance

  • Why should we care about Provenance?

  • Workflow Management Systems  

@farahzk03

https://slides.com/farahzkhan/phd-seminar

Introduction - Context

  • Workflows

  • Role of Workflows in Bioinformatics

  • Why use workflow?

  • Provenance

  • Why should we care about Provenance?

  • Workflow Management Systems  

@farahzk03

https://slides.com/farahzkhan/phd-seminar

Provenance

Information about entities, activities, and people involved in producing a piece of data or thing, which can be used to form assessments about its

Quality

Reliability

Trustworthiness

@farahzk03

https://slides.com/farahzkhan/phd-seminar

Retrospective Provenance 

The detailed record of the workflow execution including details of every process together with comprehensive information about the execution environment used to derive a specific data product.

All the details associated with a given workflow run

“Who enacted the workflow?”, “what was used to create a given data artefact?”, “when were the workflow and its processes enacted?”,  “Where was the workflow enacted?”.

 

Prospective Provenance

The ‘recipes’ used to execute a computational task, e.g. the workflow specification.

 

Workflow  Evolution

Tracking and capturing changes in workflow specifications, parameter setting, changes in the underlying software for a workflow step, or altering (adding/removing) a step

Introduction - Context

  • Workflows

  • Role of Workflows in Bioinformatics

  • Why use workflow?

  • Provenance

  • Why should we care about Provenance?  

  • Workflow Management Systems

@farahzk03

https://slides.com/farahzkhan/phd-seminar

Who did this? 

When did this happen?

Using what? 

Provenance Applications

Attribution

Quality Assurance

Verification of Results

Debugging in case of failure/error

Reproducibility

Understandability

Reuse

Trust

Introduction - Context

  • Workflows

  • Role of Workflows in Bioinformatics

  • Why use workflow?

  • Provenance

  • Why should we care about Provenance?  

  • Workflow Management Systems

@farahzk03

https://slides.com/farahzkhan/phd-seminar

 

  • Systematising the representation and management of complex computational experiments comprised of various inter-dependent data analysis tasks.

 

  • Efficient task scheduling, data management, modular methods and improved debugging. 

 

  • Automated capture of provenance information (data) to document data dependencies and the derivation process.

Workflow Management Systems (WMSs)

Examples: Taverna, Galaxy, Kepler, WINGs, GenePattern, Pegasus, Vistrails, Knime...

Problem Statement

@farahzk03

https://slides.com/farahzkhan/phd-seminar

We have workflows ... with provenance collection as a major stage of workflow life cycle

We have WMS ... with provenance capture as an important attribute of such systems

What are the expectations from published research?

@farahzk03

https://slides.com/farahzkhan/phd-seminar

Expectation?

Published workflow-centric research        end-users builds new research by exploiting sufficient provenance information including systematic methods and associated data documented in that paper.

Reality?

Different computing environment; Heterogeneous WMS; Incomplete Provenance; Limited/no access to data; Proprietary Software, Lack of standardisation, Poor understanding of provenance ...

Resulting in Workflow Decay

  1. Limited understanding of Provenance factors and the associated artefacts - Leading to incomplete provenance

  2. Heterogeneity in Workflow definition & Management - Heterogeneous Provenance capture, granularity & representation

  3. Lack of standard Representation of Workflow-centric Analysis - Results in lack of understanding &  interoperability between WMS and computing platforms

Open problems

The full "incomplete" list contains 232 entries...

And then we have many such lists.

WMS heterogeneity 

What's wrong with having so many approaches??

 Incomplete Provenance  documentation;

 

Lack of interoperability;

 

Specific Solutions to workflow design and provenance capture;

Different Provenance Representations resulting in heterogeneity

Customised research-based pipelines supporting individual scenarios

 

Vendor Lock-In;

Aim & Scope of the Study

Aim of the Study

Facilitate bioinformatics workflow-centric research understanding and improve its reproducibility and re-use  by analysing existing workflow definition approaches and identifying the fundamental elements of workflow provenance.

Building on this,

Devise techniques supporting Transparent & Complete Communication of Research.

(*) The workflow life cycle is adapted from Deelman et al. 

(**) The aggregation representation is adapted from these slides

Scope 

  • Identification of Key Artefacts of Bioinformatics Workflow Provenance - To improve the understanding and ultimately capture of the provenance

  • Conceptual Provenance Framework applicable to all - To address the heterogeneity

  • Standardised Representation of Workflow Enactments - To ensure comprehensive analysis sharing including the resources associated with provenance and achieve interoperability

Three Major Lines of Work in this Thesis

Dimensions explored:

  1. Provenance

  2. Bioinformatics Workflows

  3. Annotated & Structured Resource Aggregation

Research Questions

Research Question 1:

How are existing WMSs handling different aspects of provenance?

"different aspects of provenance"
"existing WMS"
Both classifications help in the evaluation of the existing WMSs with respect to their provenance capabilities.

Research Question 1:

How are existing WMSs handling different aspects of provenance?

"different aspects of provenance"
"existing WMS"

Provenance Taxonomy

Provenance Aspects

Provenance Supporting Resources

Research Question 1:

How are existing WMSs handling different aspects of provenance?

"different aspects of provenance"
"existing WMS"

 Workflow Approaches

Research Question 1:

How are existing WMSs handling different aspects of provenance?

"different aspects of provenance"
"existing WMS"
Both classifications help in the evaluation of the existing WMSs with respect to their provenance capabilities.

Common Trends

  1. No specific solution to provenance
  2. Dependent on execution platform for provenance capture
  3. Minimal information stored as provenance (as pdf document)
  4. Informal metadata captured
  1. Provenance elements are there but not always standardised
  2. Specific solutions such as HIstories and pages in Galaxy
  3. System specific workflows and methods
  1. Explicit declaration of resources facilitated
  2. Depends on the execution platform for provenance capture

Research Question 2: What are the key artefacts that must be documented for comprehensive provenance capture in bioinformatics workflows?

Research Question 2: What are the key artefacts that must be documented for comprehensive provenance capture in bioinformatics workflows?

The Exemplar Systems 

Variant Calling Workflow

Exemplar Workflow

Test Data (NA12878)

One from each Workflow Approach 

This work in published and implementation details are available: https://doi.org/10.1186/s12859-017-1747-0

Assumptions

What are assumptions?

Intricate details often considered needless to be documented by WMS or workflow authors

Note: Different workflow approaches have different assumptions; missing information varies

Why assumptions are not encouraged?

Assumptions result in lack of necessary details when an analysis is published

Heterogeneous & Incomplete provenance documentation

Identified Assumptions

Factors out of control of an author

  • Online third party resources
  • Learning curve of new standards

Summary

  • The intricate underlying details associated with workflow implementation, considered needless to be stated (assumed to be known)
    • Lead to various missing elements from the workflow provenance.
  • These implicit assumptions are identified through practical enactment of a complex workflow.
    • Using one exemplar system from each workflow category
  • Declarative approaches are found better for workflow definition... why??
    • Explicit nature
    • A standard is followed to specify workflows
    • Least assumptions
  • Through the implicit assumptions, key artefacts of bioinformatics workflow provenance are identified.

Conclusions

Adhering to the following proposed recommendations along with an explicit declaration of workflow specification will result in fine-grained and complete provenance capture.

Conclusions

Adhering to the following proposed recommendations along with an explicit declaration of workflow specification will result in fine-grained and complete provenance capture.

Conclusions

Adhering to the following proposed recommendations along with an explicit declaration of workflow specification will result in fine-grained and complete provenance capture.

Research Question 2: What are the key artefacts that must be documented for comprehensive provenance capture in bioinformatics workflows?

Common recommendations, best practices and standard approaches for

workflow design and workflow-centric analyses sharing

focused on improving the applications of provenance

Grouped these recommendations to identify fundamental artefacts and practices crucial for the capture of comprehensive provenance

and support the transparent sharing of workflow-centric studies.

Pragmatic Analysis of 15 prominent studies to extract 20 common recommendations

Literature 

Method 

Outcome

Findings

  • Improved understanding of

    • The crucial provenance artefacts that are required in case of bioinformatics workflows

    • Impact of these artefacts on different applications of provenance

    • Implications of incomplete documentation of provenance

Issues

  • The sharing of “all artefacts” from a computational experiment or in other words following all the recommendations and best practices arbitrarily without any informed guidance is a demanding task.
  • Again "heterogeneity"  resulting in varying
    • Provenance documentation
    • Provenance Granularity
    • Provenance representation
    • Sharing mechanisms of these artefacts

Research Question 3

How can we devise a hierarchical provenance framework encompassing community experiences that can serve as a guiding principle to determine the state of the provenance of a given published analysis designed using a particular workflow definition approach?

Issues

  • The sharing of “all artefacts” from a computational experiment or in other words following all the recommendations and best practices arbitrarily without any informed guidance is a demanding task.
  • Again "heterogeneity"  resulting in varying
    • Provenance documentation
    • Provenance Granularity
    • Provenance representation
    • Sharing mechanisms of these artefacts

Provenance Framework

Provenance Framework

Level 0 –Trust, Prospective Provenance & Reuse

Level 1 –Retrospective Provenance & Reproducibility

Level 2 –Towards White-box Enactment

Level 3 –Understandability & Specificity

Issues

  • The sharing of “all artefacts” from a computational experiment or in other words following all the recommendations and best practices arbitrarily without any informed guidance is a demanding task.
  • Again "heterogeneity"  resulting in varying
    • Provenance documentation
    • Provenance Granularity
    • Provenance representation
    • Sharing mechanisms of these artefacts

Solution

  • Hierarchical & Generic Provenance framework as a guide
  • Again "heterogeneity"  resulting in varying
    • Each level will have same provenance elements
    • The granularity of information will be similar given the level is same
    • Provenance representation - Still an issue
    • Sharing mechanisms of these artefacts - Still an issue

Summary

Research Question 4

How can we leverage existing abstraction and standardisation techniques to realise the provenance framework and demonstrate its utility?

Issues

  • The sharing of “all artefacts” from a computational experiment or in other words following all the recommendations and best practices arbitrarily without any informed guidance is a demanding task.
  • Again "heterogeneity"  resulting in varying
    • Provenance documentation
    • Provenance Granularity
    • Provenance representation
    • Sharing mechanisms of these artefacts

Solution

  • Hierarchical & Generic Provenance framework as a guide
  • Again "heterogeneity"  resulting in varying
    • Each level will have same provenance elements
    • The granularity of information will be similar given the level is same
    • Provenance representation - Still an issue
    • Sharing mechanisms of these artefacts - Still an issue

Research Question 4

How can we leverage existing abstraction and standardisation techniques to realise the provenance framework and demonstrate its utility?

CWLProv

Standardised Format for the representation of a CWL workflow enactment, associated artefacts and the retrospective provenance

Provenance using PROV-Model
expanded with wfprov and wfdesc

Workflow specifications

CWLProv Research Object

Adapted from:

https://doi.org/10.5281/zenodo.1484286#page=8

Mechanism for serialization and transport consistency

Research Question 4

How can we leverage existing abstraction and standardisation techniques to realise the provenance framework and demonstrate its utility?

Step one:

Choose a feature complete reference implementation of CWL

Ideal choice:

cwltool

  • Feature complete implementation of CWL.
  • Extensive validation of CWL files
  • Comprehensive set of test cases to validate new modules introduced as extensions to the existing implementation

Why?

CWLProv is implemented as an optional module to cwltool and can be invoked if required

Process diagram for recording provenance

cwltool --provenance ROname workflow.cwl job.json

Research Question 4

How can we leverage existing abstraction and standardisation techniques to realise the provenance framework and demonstrate its utility?

CWLProv Evaluation using workflows defined by independent research groups

Across computing platforms

Across executors

This workflow despite specified in standardised workflow definitions missed "explicit declaration"

non-deterministic algorithm: number of threads –t and the seed length –K

$ cwlprov --help
usage: cwlprov [-h] [--version] [--directory DIRECTORY] [--relative]
            [--absolute] [--output OUTPUT] [--verbose] [--quiet] [--hints]
            [--no-hints]
            {validate,info,who,prov,inputs,outputs,run,runs,rerun,derived,runtimes}
            ...

cwlprov explores Research Objects containing provenance of Common Workflow
Language executions. <https://w3id.org/cwl/prov/>

commands:
{validate,info,who,prov,inputs,outputs,run,runs,rerun,derived,runtimes}
    validate            Validate the CWLProv Research Object
    info                show research object Metadata
    who                 show Who ran the workflow
    prov                export workflow execution Provenance in PROV format
    inputs              list workflow/step Input files/values
    outputs             list workflow/step Output files/values
    run                 show workflow Execution log
    runs                List all workflow executions in RO
    rerun               Rerun a workflow or step
    derived             list what was Derived from a data item, based on
                        activity usage/generation
    runtimes            calculate average step execution Runtimes

https://slides.com/soilandreyes/2018-10-29-cwlprov#/13/2

Key Findings

  • "Explicit is better than implicit" - Tim Peters
  • Workflow abstraction can address the heterogeneity of workflow specifications to improve their interoperability across different WMS and computing platforms
    • Supported with the accompanying resources & provenance
    • CWL called “future trend” and “lingua franca”
  • Provenance capture and subsequent use to support published research transparency and integrity should not be treated as an after-thought but rather as a standard practice of utmost priority.

    • The assumption of black-box provenance often associated with the workflows

      and used to justify the coarse-grained provenance of workflow steps should not be encouraged

Summary of Contributions

  • Taxonomy Definition
  • Workflow Approaches Classification
  • Identification of Implicit Assumptions
  • A comprehensive set of recommendations
  • Identification & Understanding of Provenance Artefacts
  • Conceptual Provenance Framework

  • CWLProv –Standard Format

  • Practical Implementation to generate CWLProv ROs

  • Interoperability Demonstration across different platforms

  • Supporting Tool Development - Provenance analytics

Impact of the Research

  1. The empirical case study and its findings are already contributing in shaping recent research - https://doi.org/10.7717/peerj.5551

  2. The conceptual hierarchical provenance framework and CWLProv format are now utilised as a guide by the Nextflow team to implement research object support for the nextflow pipelines - 10.5281/zenodo.1323830

  3. A recent effort focused on extending toil-runner has commenced which will provide support for CWLProv RO generation after workflow enactment.

Thank you 

Questions?

Made with Slides.com