\n The author of the
source notebook ALEXANDER CHERVOV has proposed to transform a multi-label classification task into a binary classification of multiple targets, using protein embeddings as features and Ridge as the simplest and fastest ML model to make baseline predictions. I tried to translate his Python notebook into R, but encountered a problem related to the crazy time of processing (
9s in Python is 39 min in R).
\n The workflow is the following:\n \n
\n - Load a precalculated set of binarized dependent variables (in this case, 1499 most frequent GO terms - targets) for each protein (the presence or absence of a particular GO term) - form the prepared dataset\n
- Load precalculated embeddings for protein sequences (features) (converded from .npy to RDS files, stored in the same dataset)\n
- Use Ridge ML-model to make predictions\n
\n
Referenct notebook in Python\n
Reference Protein embeddings data\n