protein-qc

Name: protein-qc
Author: adaptyvbio

41installs20forks

How to install protein-qc?

adaptyvbio/protein-design-skills/protein-qc

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

Installs into the current project.

Summary

Quality control metrics and filtering thresholds for protein design. Use this skill when: (1) Evaluating design quality for binding, expression, or structure, (2) Setting filtering thresholds for pLDDT, ipTM, PAE, (3) Checking sequence liabilities (cysteines, deamidation, polybas

Files · 1

View on GitHub

SKILL.md

1	# Protein Design Quality Control
2
3	## Critical Limitation
4
5	Individual metrics have weak predictive power for binding. Research shows:
6	- Individual metric ROC AUC: 0.64-0.66 (slightly better than random)
7	- Metrics are pre-screening filters, not affinity predictors
8	- Composite scoring is essential for meaningful ranking
9
10	These thresholds filter out poor designs but do NOT predict binding affinity.
11
12	## QC Organization
13
14	QC is organized by purpose and level:
15
16	\| Purpose \| What it assesses \| Key metrics \|
17	\|---------\|------------------\|-------------\|
18	\| Binding \| Interface quality, binding geometry \| ipTM, PAE, SC, dG, dSASA \|
19	\| Expression \| Manufacturability, solubility \| Instability, GRAVY, pI, cysteines \|
20	\| Structural \| Fold confidence, consistency \| pLDDT, pTM, scRMSD \|
21
22	Each category has two levels:
23	- Metric-level: Calculated values with thresholds (pLDDT > 0.85)
24	- Design-level: Pattern/motif detection (odd cysteines, NG sites)
25
26	---
27
28	## Quick Reference: All Thresholds
29
30	\| Category \| Metric \| Standard \| Stringent \| Source \|
31	\|----------\|--------\|----------\|-----------\|--------\|
32	\| Structural \| pLDDT \| > 0.85 \| > 0.90 \| AF2/Chai/Boltz \|
33	\| \| pTM \| > 0.70 \| > 0.80 \| AF2/Chai/Boltz \|
34	\| \| scRMSD \| < 2.0 Å \| < 1.5 Å \| Design vs pred \|
35	\| Binding \| ipSAE_min \| > 0.61 \| > 0.70 \| AF3/Boltz (see ipsae) \|
36	\| \| ipTM \| > 0.50 \| > 0.60 \| AF2/Chai/Boltz \|
37	\| \| PAE_interaction \| < 12 Å \| < 10 Å \| AF2/Chai/Boltz \|
38	\| \| Shape Comp (SC) \| > 0.50 \| > 0.62 \| PyRosetta \|
39	\| \| interface_dG \| < -10 \| < -15 \| PyRosetta \|
40	\| \| Interface BUNS \| <= 4 \| <= 2 \| PyRosetta \|
41	\| Expression \| Instability \| < 40 \| < 30 \| BioPython \|
42	\| \| GRAVY \| < 0.4 \| < 0.2 \| BioPython \|
43	\| \| ESM2 PLL \| > 0.0 \| > 0.2 \| ESM2 \|
44	\| \| Folding ΔG \| < -2 kcal/mol \| < -4 kcal/mol \| SaProtΔG \|
45
46	### Design-Level Checks (Expression)
47	\| Pattern \| Risk \| Action \|
48	\|---------\|------\|--------\|
49	\| Odd cysteine count \| Unpaired disulfides \| Redesign \|
50	\| NG/NS/NT motifs \| Deamidation \| Flag/avoid \|
51	\| K/R >= 3 consecutive \| Proteolysis \| Flag \|
52	\| >= 6 hydrophobic run \| Aggregation \| Redesign \|
53
54	See: references/binding-qc.md, references/expression-qc.md, references/structural-qc.md
55
56	---
57
58	## Interface metrics (PyRosetta)
59
60	Beyond shape complementarity and interface_dG, two interface metrics catch common
61	de novo failure modes:
62
63	- Buried unsatisfied H-bonds (BUNS): buried polar atoms making no hydrogen bond.
64	This is a dominant energetic failure mode and is orthogonal to dG and dSASA. Keep
65	interface BUNS at or below 4 (standard) or 2 (stringent).
66	- ContactMolecularSurface: shape-complementarity-weighted contact area that, unlike
67	dSASA, is not fooled by gappy or holey interfaces. Higher is better.
68
69	Both are in the Cao 2022, AlphaProteo, and BindCraft filter sets.
70
71	For structure-quality ranking, biomodals also provides `modal_af2rank.py` (AF2Rank),
72	which scores how well a design re-predicts from its own structure as a template.
73
74	## Binder ranking (benchmark-backed)
75
76	A meta-analysis of 3,766 experimentally tested binders across 15 targets (Overath et
77	al., bioRxiv 2025, doi:10.1101/2025.08.14.670059) found that AF3 `ipSAE_min` is the
78	single best in-silico predictor of binding, and that a simple linear model of three
79	features generalizes best across targets. Complexity did not help: gradient-boosted
80	and many-feature models did not beat the linear one.
81
82	Recommended filtering strategies from that work:
83
84	1. Threshold on one of: `AF3 ipSAE_min > 0.61`, or `ipSAE_min × interface_ΔG/ΔSASA <
85	-1.5`, or `LIS × shape_complementarity > 0.42`.
86	2. Pre-filter on `shape_complementarity > 0.62` and `RMSD_binder < 3.73` (input vs
87	re-predicted), then take the top-K by `ipSAE_min`.
88
89	`ipSAE_min` is the minimum of the two asymmetric ipSAE directions (binder→target and
90	target→binder), not the average or max. Use the `ipsae` skill to compute it. Note the
91	RMSD_binder filter can be over-restrictive on some targets, so prefer it as a soft
92	pre-filter rather than a hard cutoff.
93
94	## Stability prediction (small domains)
95
96	For small domains (roughly 60 to 80 residues, the minibinder range), absolute folding
97	stability can be predicted directly. SaProtΔG (Cho et al., bioRxiv 2026,
98	doi:10.64898/2026.05.19.726285) predicts absolute fold

People who liked this also viewed

6 picks

Type

protein-qc

41installs20forks

View on GitHub

How to install protein-qc?

adaptyvbio/protein-design-skills/protein-qc

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

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Protein Design Quality Control
2
3	## Critical Limitation
4
5	Individual metrics have weak predictive power for binding. Research shows:
6	- Individual metric ROC AUC: 0.64-0.66 (slightly better than random)
7	- Metrics are pre-screening filters, not affinity predictors
8	- Composite scoring is essential for meaningful ranking
9
10	These thresholds filter out poor designs but do NOT predict binding affinity.
11
12	## QC Organization
13
14	QC is organized by purpose and level:
15
16	\| Purpose \| What it assesses \| Key metrics \|
17	\|---------\|------------------\|-------------\|
18	\| Binding \| Interface quality, binding geometry \| ipTM, PAE, SC, dG, dSASA \|
19	\| Expression \| Manufacturability, solubility \| Instability, GRAVY, pI, cysteines \|
20	\| Structural \| Fold confidence, consistency \| pLDDT, pTM, scRMSD \|
21
22	Each category has two levels:
23	- Metric-level: Calculated values with thresholds (pLDDT > 0.85)
24	- Design-level: Pattern/motif detection (odd cysteines, NG sites)
25
26	---
27
28	## Quick Reference: All Thresholds
29
30	\| Category \| Metric \| Standard \| Stringent \| Source \|
31	\|----------\|--------\|----------\|-----------\|--------\|
32	\| Structural \| pLDDT \| > 0.85 \| > 0.90 \| AF2/Chai/Boltz \|
33	\| \| pTM \| > 0.70 \| > 0.80 \| AF2/Chai/Boltz \|
34	\| \| scRMSD \| < 2.0 Å \| < 1.5 Å \| Design vs pred \|
35	\| Binding \| ipSAE_min \| > 0.61 \| > 0.70 \| AF3/Boltz (see ipsae) \|
36	\| \| ipTM \| > 0.50 \| > 0.60 \| AF2/Chai/Boltz \|
37	\| \| PAE_interaction \| < 12 Å \| < 10 Å \| AF2/Chai/Boltz \|
38	\| \| Shape Comp (SC) \| > 0.50 \| > 0.62 \| PyRosetta \|
39	\| \| interface_dG \| < -10 \| < -15 \| PyRosetta \|
40	\| \| Interface BUNS \| <= 4 \| <= 2 \| PyRosetta \|
41	\| Expression \| Instability \| < 40 \| < 30 \| BioPython \|
42	\| \| GRAVY \| < 0.4 \| < 0.2 \| BioPython \|
43	\| \| ESM2 PLL \| > 0.0 \| > 0.2 \| ESM2 \|
44	\| \| Folding ΔG \| < -2 kcal/mol \| < -4 kcal/mol \| SaProtΔG \|
45
46	### Design-Level Checks (Expression)
47	\| Pattern \| Risk \| Action \|
48	\|---------\|------\|--------\|
49	\| Odd cysteine count \| Unpaired disulfides \| Redesign \|
50	\| NG/NS/NT motifs \| Deamidation \| Flag/avoid \|
51	\| K/R >= 3 consecutive \| Proteolysis \| Flag \|
52	\| >= 6 hydrophobic run \| Aggregation \| Redesign \|
53
54	See: references/binding-qc.md, references/expression-qc.md, references/structural-qc.md
55
56	---
57
58	## Interface metrics (PyRosetta)
59
60	Beyond shape complementarity and interface_dG, two interface metrics catch common
61	de novo failure modes:
62
63	- Buried unsatisfied H-bonds (BUNS): buried polar atoms making no hydrogen bond.
64	This is a dominant energetic failure mode and is orthogonal to dG and dSASA. Keep
65	interface BUNS at or below 4 (standard) or 2 (stringent).
66	- ContactMolecularSurface: shape-complementarity-weighted contact area that, unlike
67	dSASA, is not fooled by gappy or holey interfaces. Higher is better.
68
69	Both are in the Cao 2022, AlphaProteo, and BindCraft filter sets.
70
71	For structure-quality ranking, biomodals also provides `modal_af2rank.py` (AF2Rank),
72	which scores how well a design re-predicts from its own structure as a template.
73
74	## Binder ranking (benchmark-backed)
75
76	A meta-analysis of 3,766 experimentally tested binders across 15 targets (Overath et
77	al., bioRxiv 2025, doi:10.1101/2025.08.14.670059) found that AF3 `ipSAE_min` is the
78	single best in-silico predictor of binding, and that a simple linear model of three
79	features generalizes best across targets. Complexity did not help: gradient-boosted
80	and many-feature models did not beat the linear one.
81
82	Recommended filtering strategies from that work:
83
84	1. Threshold on one of: `AF3 ipSAE_min > 0.61`, or `ipSAE_min × interface_ΔG/ΔSASA <
85	-1.5`, or `LIS × shape_complementarity > 0.42`.
86	2. Pre-filter on `shape_complementarity > 0.62` and `RMSD_binder < 3.73` (input vs
87	re-predicted), then take the top-K by `ipSAE_min`.
88
89	`ipSAE_min` is the minimum of the two asymmetric ipSAE directions (binder→target and
90	target→binder), not the average or max. Use the `ipsae` skill to compute it. Note the
91	RMSD_binder filter can be over-restrictive on some targets, so prefer it as a soft
92	pre-filter rather than a hard cutoff.
93
94	## Stability prediction (small domains)
95
96	For small domains (roughly 60 to 80 residues, the minibinder range), absolute folding
97	stability can be predicted directly. SaProtΔG (Cho et al., bioRxiv 2026,
98	doi:10.64898/2026.05.19.726285) predicts absolute fold

People who liked this also viewed

6 picks

Type