Machine Learning in single-cell RNA-sequencing analysis

Machine Learning in single-cell RNA-sequencing analysis

This is a block Seminar for Bioinformatics students. 

Overview: Machine learning has become an indispensable tool in bioinformatics, capable of extracting meaningful insights from complex biological data. Single-cell RNA-sequencing (scRNA-seq) is a powerful technique that allows for the comprehensive analysis of gene expression at the single cell level. Unlike traditional bulk RNA-seq, which provides an average expression profile of a cell population, scRNA-seq unveils the heterogeneity within a sample by capturing individual cell transcriptomes. Moreover, recent advances in sequencing technologies have led to the emergence of spatial transcriptomics (ST) which allows for the analysis of gene expression patterns within the spatial context of a tissue sample, effectively mapping gene activity to histological regions.

In this seminar, we will discuss recent publications that apply machine learning and deep learning techniques to these fields. Topics include automated cell type annotation, classification of cells, deconvolution of spots in ST data and more.

 

Dozent: Dr. Fabian Kern
Tutor: Friederike Grandke, Tobias Wolff, Matthias Flotho, Annika Engel

---

Key Dates

Seminar registration [mandatory]	From April 15 to May 13, 2024 in the CMS system
Registration for topics [mandatory]	From April 29 to May 13, 2024 in the CMS system
Kick-off meeting [mandatory]	May 22, 2024
Summary submission deadline [mandatory]	June 05, 2024
Deadline to (de-)register in HISPOS OR de-register from seminar * [mandatory]	June 12, 2024
Deadline for slide feedback (Further information can be found below)	July 22, 2024
Presentations [mandatory]	August 05 & August 06, 2024
Report submission deadline [mandatory]	August 19, 2024

 

 * If you want to de-register from the seminar, please send the tutor an email irrespectively whether you (de)registered in HISPOS or not.

 

Requirements for participation:

• Proseminar
  • Only Bachelor students
  • 5 credit points
  • Requirement: At least in the 3rd semester, passed "Bioinformatics I and II"
• Seminar
  • Master students
  • 7 credit points
  • Requirements: Passed "Elements of machine learning" (EML) or "Machine learning"

Good language skills are presumed as all talks will be held in English.
In case you want to comment on the course requirements, use the CMS Note option on your Personal Status page after registration for the course.

 

Certificate requirements:

• Summary of the assigned paper:
  • Submission until June 06, 2024
  • Approx. 3 pages of text excluding title (page), section titles and the references (at least 1500 words in Arial with standard font size 12 and single line spacing) 
  • Main structure: Title (page), main text (with or without subsections), references
  • Figures, tables and formulas are not required. Figures and tables are only accepted if they have been created completely by yourself and should be included in the 3 pages. If a figure or table is based on a figure or table in the paper, a reference to it must always be given.
  • The text should be written in your own words, no copying from the paper or other sources. The texts will be checked for plagiarism.
  • It is recommended to look for further literature to explain and classify the assigned paper.
  • It is recommended to write the summary using LaTeX in order to train scientific writing
• Successful presentation:
  • Talk: 30 minutes for a Proseminar and 40 minutes for a Seminar
  • Discussion: 10 minutes during which you should be able to answer questions from the tutor(s)/audience (All participants are strongly encouraged to actively participate in the discussion and ask questions).
  • The use of any kind of cheatsheet for the talk or the discussion is prohibited. Additionally, we point out that it is not advisable to write entire sentences in the PowerPoint note window.
  • Participation: All presentations must be attended, even if there are several presentation days.
  • Handing in a digital copy of the completed guidelines checklist and the final slides on the evening before the presentations (August 4, 23:59).
• Report of the presentations: 
  • Submission until two weeks after the presentation (August 19, 2024)
  • The report should consist of 12 paragraphs, one for each presentation.
  • Each paragraph should contain a summary of the publication and, additionnaly, it should be addressed questions and unclear points from the Q&A section of the presentation.
  • Each paragraph should be between 300 and 350 words (Arial with standard font size 12 and single line spacing).
  • The text should be written in your own words, no copying from the paper or other sources. The texts will be checked for plagiarism.
  • It is recommended to write the report using LaTeX in order to train scientific writing 

Please be advised that the utilization of language models, including but not limited to ChatGPT, for the creation of slides and/or textual content is strictly prohibited. All suspected instances will be subject to a personal interview to investigate the matter further.

 

Final grade based on

• the summary (25%) 
• the presentation & follow-up discussion (50%) 
• the report (25%)

Please note: Your slides will make up a substantial part of the final grade. Reading and paying attention to the provided presentation guidelines (see below) will help you to get an impression of which aspects are relevant for the evaluation. Disregarding many of the points listed in the guidelines may negatively affect your grade. 

 

Slides

Check out our support materials (presentation guidelines, presentation guidlines checklist). If you would like to get feedback about your slides, e.g. to improve your presentation before the talk, send your slides to the tutor before the feedback deadline (July 22, 2024). We strongly encourage you to take this opportunity. When asking for feedback, the more complete the submitted presentation is the more helpful our feedback can be. Thus, try to avoid submitting half-finished slides. Feedback will be provided at least once but at most twice per participant (if the second request was before the deadline). 

 

Topics

Paper preferences

First, you have to register for this course. Afterwards you find a registration option for each of the topics. If you are interested in one or more topics and would like to present this paper, press the corresponding register button. Multiple registrations are possible and increase the chances of getting assigned a topic. If you just register for the course and do not register for any topic, you cannot participate in the seminar and will not get a topic. Note that the course registration starts at April 15 but the topic registration opens at April 29 and closes at May 13, 2024. In this time you can register and de-register from topics as often as you want to. After the deadline, an algorithm will asign topics to students and we will invite these students to the kick-off meeting, all other students will get a rejection by mail.

Topics

All manuscript files are either open-access or available via the university network using a secure VPN connection.

Nr.	Topic	Supervised by	Type
1	Realistic in silico generation and augmentation of single-cell RNA-seq data using generative adversarial networks	Matthias	Seminar
2	scGPT: toward building a foundation model for single-cell multi-omics using generative AI	Matthias	Seminar
3	Deep generative modeling for single-cell transcriptomics	Friederike	Proseminar and Seminar
4	Transfer learning enables predictions in network biology	Matthias	Seminar
5	Phenotype prediction from single-cell RNA-seq data using attention-based neural networks	Tobias	Seminar
6	scBERT as a large-scale pretrained deep language model for cell type annotation of single-cell RNA-seq data	Annika	Seminar
7	Identifying tumor cells at the single-cell level using machine learning	Tobias	Proseminar and Seminar
8	Transformer for one stop interpretable cell type annotation	Tobias	Proseminar and Seminar
9	devCellPy is a machine learning-enabled pipeline for automated annotation of complex multilayered single-cell transcriptomic data	Friederike	Proseminar and Seminar
10	BIDCell: Biologically-informed self-supervised learning for segmentation of subcellular spatial transcriptomics data	Annika	Seminar
11	DSTG: deconvoluting spatial transcriptomics data through graph-based artificial intelligence	Friederike	Seminar
12	scPred: accurate supervised method for cell-type classification from single-cell RNA-seq data	Friederike	Proseminar and Seminar