ipsae

Name: ipsae
Author: adaptyvbio

37installs20forks

How to install ipsae?

adaptyvbio/protein-design-skills/ipsae

$npx -y skills add adaptyvbio/protein-design-skills --skill ipsae

Installs into the current project.

Summary

Binder design ranking using ipSAE (interprotein Score from Aligned Errors). Use this skill when: (1) Ranking binder designs for experimental testing, (2) Filtering BindCraft or RFdiffusion outputs, (3) Comparing AF2/AF3/Boltz predictions, (4) Predicting binding success rates, (5)

Files · 1

View on GitHub

SKILL.md

1	# ipSAE Binder Ranking
2
3	## Prerequisites
4
5	\| Requirement \| Minimum \| Recommended \|
6	\|-------------\|---------\|-------------\|
7	\| Python \| 3.8+ \| 3.10 \|
8	\| NumPy \| 1.20+ \| Latest \|
9	\| RAM \| 8GB \| 16GB \|
10
11	## Overview
12
13	ipSAE (interprotein Score from Aligned Errors) is a scoring function for ranking protein-protein interactions predicted by AlphaFold2, AlphaFold3, and Boltz1. It separates true from false predicted complexes more reliably than ipTM, which dilutes interface confidence across disordered or accessory regions. In a separate binder meta-analysis (Overath et al. 2025), AF3 ipSAE_min gave a 1.4-fold gain in average precision over the ipAE score that RFdiffusion pipelines commonly filter on.
14
15	Paper: Dunbrack, "Rēs ipSAE loquuntur: What's wrong with AlphaFold's ipTM score and how to fix it", [bioRxiv 2025.02.10.637595](https://www.biorxiv.org/content/10.1101/2025.02.10.637595v2)
16
17	## How to run
18
19	### Installation
20	```bash
21	git clone https://github.com/DunbrackLab/IPSAE.git
22	cd IPSAE
23	pip install numpy
24	```
25
26	### AlphaFold2
27	```bash
28	python ipsae.py scores_rank_001.json unrelaxed_rank_001.pdb 15 15
29	```
30
31	### AlphaFold3
32	```bash
33	python ipsae.py fold_model_full_data_0.json fold_model_0.cif 10 10
34	```
35
36	### Boltz1
37	```bash
38	python ipsae.py pae_model_0.npz model_0.cif 10 10
39	```
40
41	## Key parameters
42
43	\| Parameter \| Description \| Recommended \|
44	\|-----------\|-------------\|-------------\|
45	\| PAE file \| JSON (AF2/AF3) or NPZ (Boltz) \| Match predictor \|
46	\| Structure file \| PDB or CIF structure \| Match PAE \|
47	\| PAE cutoff \| Threshold for contacts \| 10-15 \|
48	\| Distance cutoff \| Max CA-CA distance (A) \| 10-15 \|
49
50	## Output format
51
52	Two output files are generated:
53
54	Chain-pair scores (`_chains.csv`):
55	```
56	chain_A,chain_B,ipSAE_min,pDockQ,pDockQ2,LIS,n_contacts,interface_dist
57	A,B,0.72,0.65,0.58,0.45,42,8.5
58	```
59
60	Residue-level scores (`_residues.csv`):
61	```
62	chain,resnum,pSAE,pLDDT
63	A,45,0.85,92.3
64	A,67,0.78,88.1
65	```
66
67	## Sample output
68
69	### Successful run
70	```
71	$ python ipsae.py scores_rank_001.json design_0.pdb 10 10
72	Processing design_0...
73	Found 2 chains: A, B
74	Computing ipSAE scores...
75
76	Results written to:
77	design_0_chains.csv
78	design_0_residues.csv
79
80	Summary:
81	ipSAE_min: 0.72
82	pDockQ: 0.65
83	LIS: 0.45
84	Interface contacts: 42
85	```
86
87	What good output looks like:
88	- ipSAE_min > 0.61 (primary filter)
89	- pDockQ > 0.5 (supporting metric)
90	- Reasonable number of interface contacts (20-100)
91
92	## Decision tree
93
94	```
95	Should I use ipSAE?
96	│
97	├─ What are you ranking?
98	│ ├─ Designed binders → ipSAE ✓
99	│ ├─ Natural complexes → ipTM is fine
100	│ └─ Single proteins → Not applicable
101	│
102	├─ What predictor did you use?
103	│ ├─ AlphaFold2 → ipSAE ✓
104	│ ├─ AlphaFold3 → ipSAE ✓
105	│ ├─ Boltz1 → ipSAE ✓
106	│ ├─ Chai → ipSAE (use PAE output)
107	│ └─ ESMFold → Not applicable (no PAE)
108	│
109	└─ Why ipSAE over ipTM?
110	├─ Different length constructs → ipSAE ✓
111	├─ Designs with disordered regions → ipSAE ✓
112	└─ Standard complexes → Either works
113	```
114
115	## Recommended thresholds
116
117	\| Metric \| Standard \| Stringent \| Use Case \|
118	\|--------\|----------\|-----------\|----------\|
119	\| ipSAE_min \| > 0.61 \| > 0.70 \| Primary filter \|
120	\| LIS \| > 0.35 \| > 0.45 \| Interface quality \|
121	\| pDockQ \| > 0.5 \| > 0.6 \| Supporting \|
122
123	## Batch processing
124
125	```python
126	import subprocess
127	import os
128	from pathlib import Path
129
130	def score_designs(pae_dir, struct_dir, output_dir):
131	"""Score all designs in a directory."""
132	Path(output_dir).mkdir(exist_ok=True)
133
134	for pae_file in Path(pae_dir).glob("_scores.json"):
135	name = pae_file.stem.replace("_scores_rank_001", "")
136	struct_file = Path(struct_dir) / f"{name}.pdb"
137
138	if struct_file.exists():
139	subprocess.run([
140	"python", "ipsae.py",
141	str(pae_file),
142	str(struct_file),
143	"10", "10"
144	])
145	```
146
147	---
148
149	## Verify
150
151	```bash
152	ls *_chains.csv \| wc -l # Should match number of predictions
153	```
154
155	---
156
157	## Troubleshooting
158
159	Low scores for good designs: Check PAE/distance cutoffs
160	Missing output: Verify PAE file format matches predictor
161	Inconsistent scores: Use same cutoffs across all designs
162
163	### Error interpretation
164
165	\| Error \| Cause \| Fix \|
166	\|-------\|-------\|-----\|
167	\| `KeyError: 'pae'` \| Wrong PAE format \| Check if AF2/AF3/Boltz format \|
168	\| `FileNotFoundError` \| Structure not found \| Verify file paths \|
169	\| `ValueError: no contacts` \| No interface detected \| Check chain IDs, reduce cutoffs \|
170
171	---
172
173	Next: Select top designs (ipSAE_min > 0.61) → experimental validation.

People who liked this also viewed

6 picks

Type

ipsae

37installs20forks

View on GitHub

How to install ipsae?

adaptyvbio/protein-design-skills/ipsae

$npx -y skills add adaptyvbio/protein-design-skills --skill ipsae

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# ipSAE Binder Ranking
2
3	## Prerequisites
4
5	\| Requirement \| Minimum \| Recommended \|
6	\|-------------\|---------\|-------------\|
7	\| Python \| 3.8+ \| 3.10 \|
8	\| NumPy \| 1.20+ \| Latest \|
9	\| RAM \| 8GB \| 16GB \|
10
11	## Overview
12
13	ipSAE (interprotein Score from Aligned Errors) is a scoring function for ranking protein-protein interactions predicted by AlphaFold2, AlphaFold3, and Boltz1. It separates true from false predicted complexes more reliably than ipTM, which dilutes interface confidence across disordered or accessory regions. In a separate binder meta-analysis (Overath et al. 2025), AF3 ipSAE_min gave a 1.4-fold gain in average precision over the ipAE score that RFdiffusion pipelines commonly filter on.
14
15	Paper: Dunbrack, "Rēs ipSAE loquuntur: What's wrong with AlphaFold's ipTM score and how to fix it", [bioRxiv 2025.02.10.637595](https://www.biorxiv.org/content/10.1101/2025.02.10.637595v2)
16
17	## How to run
18
19	### Installation
20	```bash
21	git clone https://github.com/DunbrackLab/IPSAE.git
22	cd IPSAE
23	pip install numpy
24	```
25
26	### AlphaFold2
27	```bash
28	python ipsae.py scores_rank_001.json unrelaxed_rank_001.pdb 15 15
29	```
30
31	### AlphaFold3
32	```bash
33	python ipsae.py fold_model_full_data_0.json fold_model_0.cif 10 10
34	```
35
36	### Boltz1
37	```bash
38	python ipsae.py pae_model_0.npz model_0.cif 10 10
39	```
40
41	## Key parameters
42
43	\| Parameter \| Description \| Recommended \|
44	\|-----------\|-------------\|-------------\|
45	\| PAE file \| JSON (AF2/AF3) or NPZ (Boltz) \| Match predictor \|
46	\| Structure file \| PDB or CIF structure \| Match PAE \|
47	\| PAE cutoff \| Threshold for contacts \| 10-15 \|
48	\| Distance cutoff \| Max CA-CA distance (A) \| 10-15 \|
49
50	## Output format
51
52	Two output files are generated:
53
54	Chain-pair scores (`_chains.csv`):
55	```
56	chain_A,chain_B,ipSAE_min,pDockQ,pDockQ2,LIS,n_contacts,interface_dist
57	A,B,0.72,0.65,0.58,0.45,42,8.5
58	```
59
60	Residue-level scores (`_residues.csv`):
61	```
62	chain,resnum,pSAE,pLDDT
63	A,45,0.85,92.3
64	A,67,0.78,88.1
65	```
66
67	## Sample output
68
69	### Successful run
70	```
71	$ python ipsae.py scores_rank_001.json design_0.pdb 10 10
72	Processing design_0...
73	Found 2 chains: A, B
74	Computing ipSAE scores...
75
76	Results written to:
77	design_0_chains.csv
78	design_0_residues.csv
79
80	Summary:
81	ipSAE_min: 0.72
82	pDockQ: 0.65
83	LIS: 0.45
84	Interface contacts: 42
85	```
86
87	What good output looks like:
88	- ipSAE_min > 0.61 (primary filter)
89	- pDockQ > 0.5 (supporting metric)
90	- Reasonable number of interface contacts (20-100)
91
92	## Decision tree
93
94	```
95	Should I use ipSAE?
96	│
97	├─ What are you ranking?
98	│ ├─ Designed binders → ipSAE ✓
99	│ ├─ Natural complexes → ipTM is fine
100	│ └─ Single proteins → Not applicable
101	│
102	├─ What predictor did you use?
103	│ ├─ AlphaFold2 → ipSAE ✓
104	│ ├─ AlphaFold3 → ipSAE ✓
105	│ ├─ Boltz1 → ipSAE ✓
106	│ ├─ Chai → ipSAE (use PAE output)
107	│ └─ ESMFold → Not applicable (no PAE)
108	│
109	└─ Why ipSAE over ipTM?
110	├─ Different length constructs → ipSAE ✓
111	├─ Designs with disordered regions → ipSAE ✓
112	└─ Standard complexes → Either works
113	```
114
115	## Recommended thresholds
116
117	\| Metric \| Standard \| Stringent \| Use Case \|
118	\|--------\|----------\|-----------\|----------\|
119	\| ipSAE_min \| > 0.61 \| > 0.70 \| Primary filter \|
120	\| LIS \| > 0.35 \| > 0.45 \| Interface quality \|
121	\| pDockQ \| > 0.5 \| > 0.6 \| Supporting \|
122
123	## Batch processing
124
125	```python
126	import subprocess
127	import os
128	from pathlib import Path
129
130	def score_designs(pae_dir, struct_dir, output_dir):
131	"""Score all designs in a directory."""
132	Path(output_dir).mkdir(exist_ok=True)
133
134	for pae_file in Path(pae_dir).glob("_scores.json"):
135	name = pae_file.stem.replace("_scores_rank_001", "")
136	struct_file = Path(struct_dir) / f"{name}.pdb"
137
138	if struct_file.exists():
139	subprocess.run([
140	"python", "ipsae.py",
141	str(pae_file),
142	str(struct_file),
143	"10", "10"
144	])
145	```
146
147	---
148
149	## Verify
150
151	```bash
152	ls *_chains.csv \| wc -l # Should match number of predictions
153	```
154
155	---
156
157	## Troubleshooting
158
159	Low scores for good designs: Check PAE/distance cutoffs
160	Missing output: Verify PAE file format matches predictor
161	Inconsistent scores: Use same cutoffs across all designs
162
163	### Error interpretation
164
165	\| Error \| Cause \| Fix \|
166	\|-------\|-------\|-----\|
167	\| `KeyError: 'pae'` \| Wrong PAE format \| Check if AF2/AF3/Boltz format \|
168	\| `FileNotFoundError` \| Structure not found \| Verify file paths \|
169	\| `ValueError: no contacts` \| No interface detected \| Check chain IDs, reduce cutoffs \|
170
171	---
172
173	Next: Select top designs (ipSAE_min > 0.61) → experimental validation.

People who liked this also viewed

6 picks

Type