—- TO INVESTIGATE FURTHER —- #### OVERALL #### OVERALL

Intro Topics

  • outline the rough product that you are aiming for, an integrative platform that has a generally applicable library in the backend and a visualization platform.
  • it’s important to index all avaialble data in a way that is easily accessible and searchable
  • would be nice for the method of indexing to be future-proof/extendable to new datasets.
  • enumerate the datatypes/databases that you’ll be speaking about to draw some boundaries

Sequence Block

Isotypes characterization by species & cell types PTMS characterization by species, cell types and isotypes

  • Start with broad family HMMs (alpha, beta, etc.) For each family.
    • Collect all matching sequences
    • Perform clustering (try several thresholds)
    • Build phylogenetic trees to identify natural groups
    • Extract C-terminal regions and cluster them separately
    • Compare these different clustering results

Where clusters agree across methods, define these as strong sub-families Build HMMs for these sub-families Test with known examples to evaluate discrimination power Iteratively refine as needed

The above for families of tubulins and, separately, for families of MAPs. Stathmin-lke

  • PTMs possibly included into clustering/families/searches via ProForma Notation
  • search/discovery in a single modality instead of searching across various dbs, fasta files, mass spec records etc.

Helices as domain “views” into families.

Control Layer

  • introduce neo4j graph database as a way to keep track of semantic data and connect it to the structural models

Structural Domain

Index PDB structures and incorporate them into the graph according to the sequence block.

  • introduce capabilities of molstar as the the visualizing applications.
  • Raise interactivity-vs-generality tradeoff
  • search/navigation/visualization/comparison of the above
  • comparison/alignment (across what? sequences)
  • ligand binding sites

Then go more indepth to augmneting each of the individual following aspects of the data at scale:

Creating 3D structures of proteins from from sequences via AlphaFold

PTM Reconstruction workflows

Reconstructing PTMs on top of a template structure via:
  - Rosetta/
  - PSIptm

Ligand Binding Sites

Fragments 
Ligands Classification via 
Binking Pokets, Sites

Applications for Model building:

- HMM families-based deep learning models for automating CryoEM model building :
    https://www.nature.com/articles/s41586-024-07215-4
    https://www.biorxiv.org/content/10.1101/2025.03.16.643561v1
    https://www.biorxiv.org/content/10.1101/2024.11.13.623164v1.full.pdf

MD applications