Clarify relevance of SSL methods to downstream tasks in related domains

Establish a clear understanding of how self-supervised learning pre-training methods in related scientific domains—such as single-cell RNA-seq (e.g., scBERT), antibody language modeling, and protein geometric structure pretraining—relate to and benefit their respective downstream tasks.

Background

While this work analyzes and connects molecular self-supervised pre-training methods to specific categories of downstream tasks, the authors note that in adjacent domains (single-cell RNA-seq, antibodies, proteins) the understanding of self-supervised methods and their relevance to downstream tasks remains unclear.

They highlight this as a broader open direction to extend the methodological-task alignment beyond molecules to other bio-related modalities.