MORPH (Missense on Reference Protein Human)

Human missense variation, mapped onto protein structure.

MORPH (Missense on Reference Protein Human) connects gnomAD population frequencies and human protein structure predictions.

The guided notebook(s) are the intended starting point for those who want to reproduce visuals for their protein of interest with little code knowledge.

The Python and shell scripts support users who want to run jobs on their machines, rebuild the underlying datasets, or do proteome level analyses.

What you can explore

• An optional structure prediction: includes a Boltz-2 notebook when you need to generate an input structure.
• Population variation in structural context: map gnomAD missense allele frequencies onto a reference protein prediction and explore where observed missense variants occur in 3D.
• Differences across ancestry groups: build category-specific views using the frequency columns available in the prepared data.
• Alternative side chains: Inspect alternative amino-acid side chains alongside the reference structure at a glance in PyMOL using pyMUT-generated chains.

Choose your workflow

Your goal	Where to start	Experience needed
Generate an input structure with Boltz-2		Google Colab GPU runtime [1-Boltz2.ipynb]
Validate and map variations		Google Colab CPU runtime [2-MORPH.ipynb]
Walk through of PyMOL visualization	3-PyMOL.ipynb	PyMOL
Map prepared variant data on your machine	Advanced usage guide	Python, command line, and environment setup
Rebuild or extend the prepared datasets	Data-generation pipeline	Bioinformatics, Hail, Ensembl VEP, and Linux/HPC workflows

Prepared data

Public dataset: Zenodo record 18163874.

File	Purpose
lookup_amissenseXgnomAD.tsv	Maps each available UniProt accession to an archive and its per-protein TSV
amissenseXgnomAD_part*.tar.gz	Batches of per-protein variant tables combining AlphaMissense data, gnomAD frequencies, and VEP/ClinVar annotations

What the outputs represent

Mapping	Output	Meaning
Overall gnomAD frequency	<UniProt>_AF_gnomad.pdb	Frequency bins on the reference chain, with up to three ranked variant chains (V, W, X)
Selected gnomAD categories	<UniProt>_category_gnomad.pdb	One variant chain per selected category; chain/category assignments are recorded in PDB remarks

The mapped PDB B-factor field stores visualization values. For gnomAD outputs these are frequency bins. Keep the original structure if you need its original B-factors or model-confidence values.

Variant chains combine substitutions across positions to help display alternative side chains. They are visualization constructs, rather than experimentally determined mutant proteins or observed haplotypes. A position without retained frequency data should not be interpreted as proof that variation is absent.

See the advanced guide for the bin scheme, chain behavior, and current implementation limitations.

Repository layout

README.md                         Project overview and notebook entry point
LICENSE                           Project license
1-Boltz2.ipynb                    Boltz-2 structure prediction notebook
2-MORPH.ipynb                     Variation mapping notebook
3-PyMOL.ipynb                     PyMOL visualization walkthrough
docs/advanced-usage.md            Command-line workflows and annotation details
requirements/                     Development environment definitions and exports
example_pdbs/                     Example mapped PDB structures
src/
  1-data_generation/              Download, harmonize, annotate, and package data
  2-map_2_structure/              Map prepared values and substitutions to PDBs
  3-PyMOL/                        PyMOL coloring and selection helpers
data/                             Local raw and processed data (ignored by Git)
log/                              Local execution logs (ignored by Git)

Sources and acknowledgements

MORPH builds on the following datasets and tools:

• AlphaMissense: Cheng et al., Accurate proteome-wide missense variant effect prediction with AlphaMissense, with precomputed predictions on Zenodo.
• gnomAD: population variant frequencies.
• UniProt: reference protein sequences and metadata.
• Boltz-2: optional biomolecular structure prediction.
• pyMUT: rotamer-based side-chain mutation.
• Ensembl VEP and ClinVar: annotations used by the data-generation pipeline.
• Tamas Hegedus, hegelab.org: author credited in the supplied AlphaMissense PyMOL coloring helper.

When reporting results, cite the upstream datasets and tools used, and the public MORPH Zenodo record linked above.

License

Unless otherwise noted, original MORPH materials in this repository are licensed under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).

This includes the project documentation, notebooks, and other original materials.

Third-party software, datasets, models, and dependencies retain their own licenses and attribution requirements.