Hi everyone! Firstly, a huge thank you to all who contributed to the nf-core/proteinfamilies pipeline—it’s been extremely useful for my research. I'm currently studying leucine-rich repeat (LRR) containing proteins in brown algae. I'm looking to gather opinions on an idea to explore structural diversity (e.g., different domain architectures) and relationships between protein families (identified by the pipeline) by identifying shared "pseudo-domains" (for lack of a better term). Context: The pipeline identifies several LRR protein families. InterProScan analysis on homologous regions from the family MSA has shown known domains (e.g., LRR) alongside significant unannotated regions (~500-600 amino acids). I'm particularly interested in examining how these domains and conserved, unannotated regions are shared or vary across different protein families. Proposed approach: - Fragment input protein sequences into smaller segments. - Run the pipeline on these fragments to cluster similar segments and create HMM profiles. - Apply these fragment-based HMM profiles back onto full protein sequences to identify relationships and structural variations via shared "pseudo-domains." Questions so far (if it makes sense): - What’s the best way to select these fragments. Should I use a sliding window or base them on coordinates of known domains? - Would incorporating existing HMM profiles (e.g., Pfam) during the hmmsearch step improve this approach? If any of this sparks ideas, I'd greatly value your insights. Thanks a lot! Joan Lluis

Hi Joan! I am very happy that you are getting value from running proteinfamilies! A few things I've spotted, and we can start from there. 1. You'll definitely also find use in #C084HP11VPF at some point, to try and discover as much as possible for your unannotated parts 2. For the fragmentation, you could potentially use software such as TED, or chainsaw (I think they work with structural input) to fragment your sequences to the minimum viable domain fragments. If we could find software that can do this ok the sequence level, we could add an optional preprocessing step on the pipeline 3. You can always predict structures with #C0351A8S487 then chop, and then come back to proteinfamilies otherwise :D

Hi! Thanks a lot for these helpful suggestions. I'll explore the options you mentioned, and also look into other available tools for sequence-level fragmentation. I'll post if something interesting comes fromt his. Cheers!