Pro-FSFP: Few-Shot Protein Fitness Prediction
In their paper The wuthors show tthe ability to use a meta-model that is able to train models using a 'meta learning model that works with multiepl tasks to create a meta-learned model (PLMS with LORA adapters) to create better results using a ranking loss. Comparing in this manner allows for multiple results in different experiments to be used simultaneously without impacting the quality of results.