Folding Principles Dataset – ββ‑Hairpin Prototype (WIP)

Extracts β‑hairpin (ββ‑unit) motifs from PDB mmCIF structures and analyses the relationship between loop length and chirality (L / R), reproducing the trend first reported in Koga et al. 2012, Nature 491:222‑227.

Only the ββ‑pattern is implemented for now. The remaining Koga motifs (αα‑corner, βαβ‑unit, αβ‑motif, etc.) will be added in later commits.

---

1 · Environment (set‑up once)

git clone https://github.com/your‑name/FoldingPrinciplesDataset.git
cd FoldingPrinciplesDataset

# (x86 Linux / macOS Intel)
conda env create -f environment.yml
conda activate folding-dssp

# (macOS Apple Silicon) – the dssp wheel is x86 only
export CONDA_SUBDIR=osx-64
conda env create -f environment.yml
conda activate folding-dssp
conda config --env --set subdir osx-64

2 · Run the notebook

The current workflow is demonstrated in notebooks/test_pipeline_parts.ipynb.

1. Open the notebook after activating the Conda environment.
2. Edit the pdb_ids cell (or leave it empty to let the PISCES fetcher choose high‑quality structures automatically).
3. Run all cells. The notebook will
   • download the required mmCIF files
   • run DSSP on each structure
   • detect β‑hairpins
   • compute chirality with both methods
   • produce the loop‑length × handedness bar chart

3 · Dataset structure

idx	PDB	Chain	strand1_start	strand1_end	strand2_start	strand2_end	loop_len	handedness	handedness_mag	FullChainSequence	HairpinSequence	LoopSequence
0	3vor	A	69	70	75	76	4	L	−0.555	GSDSRTVSE…	RNGISGDY	GISG
1	3vor	A	75	76	78	83	1	L	−0.622	GSDSRTVSE…	DYIGIGGAI	I
2	3vor	A	78	83	93	101	9	L	−0.681	GSDSRTVSE…	GIGGAITSSGSTINKGFAMELNGL	TSSGSTINK
3	3vor	A	138	139	149	151	9	L	−0.786	GSDSRTVSE…	VNMLAATDNTTILR	MLAATDNTT
4	5gji	A	354	359	368	374	8	L	−0.229	SNMSLQNAE…	TFLVRDASTKMHGDYTLTLRK	ASTKMHGD
5	5gji	A	368	374	377	386	2	L	−0.244	SNMSLQNAE…	YTLTLRKGGNNKLIKIFHR	GG
6	5gji	A	377	386	389	391	2	L	−0.420	SNMSLQNAE…	NNKLIKIFHRDGKYG	DG
7	5gji	A	389	391	398	399	6	L	−0.275	SNMSLQNAE…	KYGFSDPLTFS	FSDPLT
8	4nsv	A	8	9	25	30	15	R	0.415	GVSGSCNID…	IDVVCPEGNGHRDVIRSVAAYSR	VVCPEGNGHRDVIRS
9	4nsv	A	25	30	33	41	2	R	0.956	GVSGSCNID…	VAAYSRQGTMWCTGSLV	QG

4 · Plot of loop length versus handedness for all 11,116 high-quality pdb entries

Local method of loop chirality definition was used.

Additional filter used on abs(handedness_magnitude) > 0.75

Appendix · Exact vs Backbone ββ‑chirality methods OUTDATED!

Function (in dataset/motif_logic.py)	Vector u	Vector v	Vector n	Best for
detect_hairpins_and_chirality (“exact / strand‑axis”)	First‑strand axis  (Cαend − Cαstart)	Midpointstrand1 → Midpointstrand2	Cα→Cβ of residue before the loop	Fast; matches most published β‑hairpin surveys
detect_hairpins_and_chirality_backbone (“backbone”)	Backbone N→C of residue before loop	Cαpre → Cαpost (across the loop)	Same Cα→Cβ side‑chain vector	Robust near the loop; insensitive to global sheet twist

Both return two chirality columns:

• handedness → “L” or “R” (sign of the scalar triple product)
• handedness_magnitude → |scalar triple product|, normalised, ∈ [-1 … 1]

A magnitude close to 0 indicates an almost planar ββ‑unit where handedness is ambiguous. Perhaps we should keep only “strong” events: