-"Feature-learning in Short Linear Motifs (SLiMs) and Characterizing SLiMs in their Structural Context."
- Created a convolutional neural network which infers a probability distribution of residues along alternative sequences of SLiMs, from the input of their atomistic structure.
- 8 week project.
- Intended as a learning exercise in machine learning.
Main file = SLiMsNN working.py
Conclusions: Working with limited accuracy, strong proof of concept, requires hyperparameter and model tweaking beyond what is avaialble in a limited timeframe.
Implicaitons: Part of the continued research of Dr. Wouter Krogh Boomsma In machine learning for structural biology informatics.
Credits: 'deepfold' scripts were created by Dr. Wouter Krogh Boomsma (Uni of Copenhagen) and associates.