Hello! Here's the code and data to play around with models for the Deep Learning module of the SysBio 2026 course at EMBL-EBI. The coding task involves inferring transcription sequence motifs from experimental data. This is an adaptation of the tutorial orginally from A primer on deep learning in genomics. Key changes include switching from TensorFlow to PyTorch to illustrate the weeds of how models are trained, and adding linear and multi-layer perceptron models for comparison.
- Install the conda environment in
environment.ymlusing conda or mamba (skip this if you're using a virtual machine) - Activate the environment and launch
jupyter labin the terminal (skip this if you're using a virtual machine) - Run the notebook
CNN_motif_pytorch.ipynb
- How convolutional neural networks and multi-layer perceptrons are defined in PyTorch
- What model training involves
- Examine what convolutional networks learn after training
- Compare performance of different models at predicting TF binding and motif detection