Protein Prediction II for Bioinformaticians

What it is about

In this module you will learn modern methods for predicting protein function from sequence and structure and applying them. You understand which aspects of function can be predicted, which data and ML strategies are required for that, and how to avoid overfitting through data leakage and bias. In the end you can design, evaluate, and implement your own prediction approaches down to pseudocode level.

What you will be able to do

• Understanding of the central concepts of protein function prediction
• Application of modern ML/AI methods to function predictions
• Recognizing and avoiding overfitting, data bias and data leakage
• Use of evolutionary information (MSAs) and the limits of their use
• Classification and evaluation of protein language models (pLMs)
• Development and evaluation of your own prediction tools in team work
• Critical analysis and peer review of published methods
• Communication of the strengths and limits of predictions to experimentalists

What the module consists of

• LecturesProvision of the theoretical foundations for sequence- and structure-based function predictions as well as ML concepts
• SeminarsInteractive discussions on current methods and studies
• ExercisesPractical implementation and application of prediction methods; data preparation and evaluation
• Project work (group)Development or testing of an ML-based solution with subsequent presentation

Teaching method

• Lecture (interactive seminars, projector and whiteboard)Introduction and deepening of the theory; parts explained on the whiteboard; recordings and slides available
• Exercises and group projectsPractical implementation of lecture content and performance evaluation of own tools
• Seminars/PresentationsPresentation and discussion of project results in front of fellow students and tutors