Draft curriculum: Overview session, optional masterclasses and suggestions

[malvikasharan]

Summary

Worked out by Lydia, Fede, Malvika

Considering all the comments from James Ben McArthur, B. and James F., we have come up with a proposal to design an overview masterclass (compulsory), followed by 4 optional masterclasses as described below:

Overview Masterclasses:

• Managing open and reproducible computational Projects: challenges and benefits (reproducibility, grants, peer review, collaboration, open tools, open science)
• Introduction to Machine Learning, Deep Learning, AI (including considerations from the PIs)

Open & Reproducible Pipelines:

• Part 1: From raw data to computational analysis
• Part 2: From computational analysis to publication/sharing

Management of Computational Projects:

• Part 3: Data, code, and research objects
• Part 4: Managing teams and collaborations

--- Initial plan - revised above --- Overview Masterclass:

• Covering challenges and benefits (reproducibility, grants, peer review, collaboration, open tools, open science)

Open & Reproducible Pipelines:

• Part 1: From raw data to computational analysis
• Part 2: From computational analysis to publication/sharing

Management of Computational Projects:

• Part 3: Data, code, and research objects
• Part 4: Managing teams and collaborations

Machine Learning, Deep Learning, AI

• Part 5: Introduction to these topics including considerations from the PIs

[malvikasharan]

@LydiaFrance should the part 5 be an (second) overview masterclass?

[LydiaFrance]

@malvikasharan I think it can be thought of as separate to the rest, yes. It would be an overview of deep learning/AI and associated problems.

[malvikasharan]

Meeting notes from 03 December 2021:

• Plan the agenda for 9 Dec
• Get some feedback from Crick on the AI/ML masterclass
  • Identify the motivation, intention, learning objective
• See how we involve Sebastian A. and the PDRAs
• Focus group for next year with the right target audience
• Mention what we offer at Turing, Carpentries and Crick (ask them what they would like to mention) from bottoms up and how they can be connected with these masterclasses
• Have an agenda point for Rebecca to report on: "Crick will be consulting with a range of our Group Leaders and operational teams regarding training needs at a meeting in early December. We will be able to share more detailed information after that date."

[malvikasharan]

Draft agenda points:

• Introduce new members
• Update on the curriculum design by Lydia & Fede (draft here: https://github.com/alan-turing-institute/data-training-for-bioscience/issues/17)
• Update from Rebecca on expectationa/needs from this project for Group Leaders and operational teams
• Open dicussion on motivaion, learning objectives and next steps to include specifically in in AI/ML

[malvikasharan]

Revised target audience, objectives (with feedback from the Crick researchers)

Target audience

Experimental biologists and biomedical research communities, with a focus on two key professional/career groups

1. Senior Group leaders without any prior experience with Data Science and ML/AI - interested in understanding the potential additionality and application in their areas of expertise
2. Post doc/Lab scientists – next generation senior leaders - interested in additionality, but also the group more likely to benefit from tools to equip them with the requirements to enable the integration of computational science into biomedical science

Content

Two masterclasses will provide an overview on the potential additionality of AI/ML to life science disciplines, and to build a shared understanding of good practice principles to facilitate the integration and reproducibility of computational data science, into these areas.

Masterclasses

• Part 1: Managing open and reproducible computational Projects: challenges and benefits (reproducibility, grants, peer review, collaboration, open tools, open science)
• Part 2: Introduction to Machine Learning, Deep Learning, AI (including considerations from the PIs) ← integrating feedback from Crick, see below

Learning objectives

At the end of these masterclasses, attendees:

• will understand what successful applications for open and reproducible Data Science and AI/ML approaches in life sciences looks like
• will be able to examine challenges and potential pitfalls in applying best practices Data Science and AI/ML approaches (mismatch between ML method and problem; reproducibility, software sustainability, reproducibility, grants, peer review, collaboration, open tools, how to find expertise, open science)
• design computational projects using collaborative, reproducible and open science approaches (for example building ML image tools for cells on GitHub, developing deep learning algorithm for genomics by involving multiple groups)
• explore what additionality data science and ML/AI can bring to their research
• identify tools and practices required for managing computational projects
• assess resource required for them to effectively apply data science and AI in their context
• apply systematic framework to introduce computational protocols in their groups

Next steps for learners

• Connecting training materials with set of resources for deeper dive into various topics required to learn and apply Data Science, AI and computational research best practices
• Establishing connections with other training and training materials offered by the Turing and the Crick
• Linking with other projects in open research, data science and AI that offers opportunities to develop/enhance technical skills

Next stage

• Start with UX (as proposed in the original charter): 1:1 or focus group with the Crick PIs (potential target audience): Alden to check if RAM can support this work
• Develop masterclasses, apply examples from the Turing and Crick (look for these examples in the UX process)

----Feedback from Crick Researchers----

1. Foundational: Introduction to machine learning including:

• Intro to different algorithm types and how to choose what to use
• Splitting datasets appropriately and avoiding leakage.
• Problem of imbalanced data sets and of overfitting
• Statistics to evaluate output (precision-recall etc)
• Pointers to where to start with different approaches

2. Specific/specialist topics

3. Computational project management/supervising computational projects

The underlying theme for all this is how to make computational projects transparent, reproducible etc. The goal would be a framework for a GL to feel more confident supervising a computational biologist. This course could include:

• Organising a project: Practical directory and file structures for comp projects
• Metadata, metadata…. Use relevant examples to the Crick (e.g. genomics and imaging)
• Intro to Github and importance of version control
• Intro to markdown and Jupyter notebooks