carpentries-incubator / data-science-ai-senior-researchers Goto Github PK

Many of the ideas behind ML can also be seen in the ideas behind traditional stats, so it would be good to explain why the apparent distinction exists and why you might be drawn to neural network over a student's t-test (and vice-versa).

• Add a more detailed case study for simulation at the Turing (and the Crick?)
• Use the callout formatting for the case study
• Adding images to improve aesthetics (not critical)

Summary

• Here I have collated all referenced initially compiled in https://github.com/alan-turing-institute/data-training-for-bioscience/issues.
• Please do comment below if we have not included some resources that should be added.

References:

Turing resources

• The Turing Way materials: https://the-turing-way.netlify.app/welcome
• RSE4DataScience: https://github.com/alan-turing-institute/RSE4DataScience18
• https://alan-turing-institute.github.io/rds-course/modules/m1/1.1-WhatIsDataScience.html
• RDS course - under development: https://github.com/alan-turing-institute/rds-course
• Data Science Skills: https://github.com/alan-turing-institute/DataScienceSkills
• RSE Summer School: https://github.com/alan-turing-institute/DH-RSE-Summer-School

Other resources

• RDA-CODATA Summer School: https://github.com/CODATA-RDA-DataScienceSchools (Check with Patricia H. and DCC)
• EMBL-EBI resources: https://competency.ebi.ac.uk/ -> The UX aspect of this resource is also quite great.
  • https://twitter.com/EBItraining/status/1427197636730306563?s=20
• Open, Data-driven Science for Decision Makers
• https://training.csx.cam.ac.uk/bioinformatics/course/bioinfo-python/
• Data science in Python: https://training.csx.cam.ac.uk/bioinformatics/event/3748813
• API and biocuration: https://github.com/malvikasharan/2019-Biocuration-Module-2
  • API: EBI version of databases - proteins
  • Features: localization, pathways, intact (do two proteins interact?)
• Also check resources from: https://softwaresaved.github.io/python-intermediate-development/
• Foundations of Open Science Skills 2021
• Markowetz, F. (2017). All biology is computational biology. PLoS Biol., 15(3), e2002050. doi: 10.1371/journal.pbio.2002050).
• EOSC-Pillar RDM training catalogue, potentially anything here is relevant (The Turing Way is listed there)

Carpentries Resources

• Genomics Lessons: https://datacarpentry.org/lessons/#genomics-workshop
• Ecology Lessons: https://datacarpentry.org/lessons/#ecology-workshop

Going through the list of current incubators, here a a few projects that deal with ML already that we should connect with and reuse

• https://carpentries-incubator.github.io/ml4bio-workshop/
• https://carpentries-incubator.github.io/machine-learning-novice-python/
• https://carpentries-incubator.github.io/machine-learning-novice-sklearn/
• http://datascibc.org/Data-Science-for-Docs/
• https://carpentries-incubator.github.io/machine-learning-librarians-archivists/
• https://carpentries-incubator.github.io/bioc-intro/

Packaging and publishing

• https://carpentries-incubator.github.io/introduction-to-conda-for-data-scientists/
• https://carpentries-incubator.github.io/python-packaging-publishing/

different levels of cloud computing and HPC training materials

• http://www.hpc-carpentry.org/
• https://carpentries-incubator.github.io/hpc-intro/

AI and Machine Learning

• https://www.nature.com/articles/nrg3920
• https://turing-commons.netlify.app/
• https://medium.com/coders-mojo/quick-recap-60-days-of-data-science-and-machine-learning-series-457ca0e98279?sk=c681bf663d6a0311454a43f755e3ed87
• https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6616181/
• https://pubmed.ncbi.nlm.nih.gov/34398394/
• https://www.astrazeneca.com/r-d/data-science-and-ai.html
• https://www.ddw-online.com/ground-breaking-techniques-in-the-use-of-ai-to-predict-cell-growth-8646-202012/
• https://www.nature.com/articles/s41524-019-0221-0

• Adding context to the overview of the training materials (Line 54)
• Add Learning Outcomes Section
• Some repetition with Modular and Flexible Learning Section with Mode of Delivery -- combine?
• A section on what the course will not cover?

I have put placeholder images i.e. a ROC curve that has been sourced from a blog / site and maybe it would be better if a custom illustration might work better? We could produce these via an example analysis we conduct ourselves (and link a notebook) if that doesn't sound the way to go? Just thinking about consistency.

For example here is a placeholder image I am using in a PR here

Supervised Classification - cover accuracy, true positives rate, false positive rate, precision, recall, ROC.
Unsupervised - kmeans total within sum of squared error, elbow
Regression - RMSE

Hi my name is Jo, some of you know me already from other co-creation projects and activities around Open Science.
It's a great honor to join the team and impressive to see what has been built already.

Moving forward with this and the "data-science-ai-senior-researchers repository, I'll go through the modules to make suggestions (pull requests and issues), as agreed with @malvikasharan and @jcolomb in order to finalize the content and to get both courses ready by mid-April, provided that we all agree.

Thanks for all your excellent contributions thus far. Both courses will surely be of great value to many researchers around the world.
Ping me with any comments or concerns you might have, I am looking forward to working with you.
Best wishes
Jo.

Incorporate case studies or relevant examples of ML/AI applications from your work into a lesson either by editing the raw files, of use a HackMD or Google doc to create your draft and share by replying below.

Also check if your examples fit better in the training for Data Science and AI introduction: carpentries-incubator/managing-computational-projects#42

Present the common metrics and visualizatons that are used to report machine lewrning results i.e. confusion matrix, ROC curve, elbow plot etc.

In the introduction I think we should contextualize the use of AI and it's relation to ML. AI is a very high level term that gets used less as you read papers on methodologies etc where ML and sub-types of ML are more common. We should explain why this is.

• Use correct formatting for the case studies (test the images and gifs)
• Expand on the benefits to open tool communities and impact
• Possibly refactor the episode into learning-outcomes.
• Are there automation tools used by the Crick?

This lesson should cover foundational concepts introducing

• data science, machine learning, Deep Learning, AI
• Address ‘why do these apply to me?’
• Include examples to inform future directions in biomedical and related fields: breakthrough technology and applications
• 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
• Guide to common tools - perhaps this will be covered in the above sections, but a guide to available commonly used tools would be helpful.

Examples should be pulled from the Turing where we can and also more broadly from international communities highlighting specific/specialist topics

• Genomics: Single cell genomic analyses (scRNAseq, scATACseq, spatial transcriptomics);
• Genomics: Cancer genomics
• Imaging: Biological image analysis (ML for segmentation, tracking, registration, classification, de-noising).
• AlphaFold, protein folding- structural biology/protein-protein interaction
• See some more examples here: carpentries-incubator/managing-computational-projects#19 (comment)

Another lesson is being developed alongside for Managing open and reproducible computational projects: challenges and benefits (reproducibility, collaboration, open tools, version control and importance in grants, peer review, academic incentives)

• https://carpentries-incubator.github.io/managing-computational-projects/