Lacuna: Cryptic Binding Pocket Discovery via Conformational Ensemble Analysis
bioinformatics
Quick verdict
Open-source ensemble tool for cryptic pocket discovery; honest, well-ablated evaluation (55.6–73.9% CryptoBench) but test fold was seen during development.
Summary
Lacuna is an open-source Python tool that finds cryptic binding pockets — cavities absent or too small to detect in a protein's resting structure — by generating a conformational ensemble, detecting pockets in each conformer, clustering detections into persistent sites, and ranking them with a linear model over 23 features fitted on within-structure pairs. The default ensemble comes from anisotropic-network-model normal modes and completes in a median of 2.6 seconds per chain on one CPU core; implicit-solvent MD, Boltz-2, and user-supplied ensembles are pluggable alternatives. On the CryptoBench test fold (180 structures), top-five recovery is 55.6% by default and 66.1% with an optional protein-language-model ranker, against 43.9% for MDpocket given the identical ensemble; pooling geometric and learned-surface detectors raises recovery to 73.9% and site coverage to 86.4% in a separate five-conformer evaluation. External results are 73% on PocketMiner, 45% on curated literature apo/holo pairs, and 86.8% on the COACH420 control, where P2Rank leads by 6.9 points. The authors candidly report that the default trails P2Rank on cryptic sites by 7.8 points and that ranking, not detection, limits performance. Key limitations: test-fold performance was observed during development (held-out but not blind), fusion results use a different evaluation protocol from the headline numbers, and there is no direct comparison with PocketMiner despite using its dataset.
Review complete
Assessment at a glance
5
risk flags identified
● high 0
● medium 3
● low 2
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Cheap elastic-network ensembles combined with per-conformer detection and cross-conformer clustering make ensemble-based cryptic-pocket discovery routine, at a median of 2.6 seconds per chain on one CPU core.
- A fitted linear ranking model over 23 features more than triples CryptoBench test-fold recovery versus the analytic crypticity rule it replaced (17.8% to 55.6%) on identical candidates.
- Lacuna recovers 55.6% of CryptoBench test-fold sites in its top five (66.1% with PLM features), exceeding MDpocket given the identical ensemble.
- Pooling geometric and learned-surface detectors raises top-five recovery to 73.9% and site coverage to 86.4% in a separate five-conformer held-out evaluation.
- Ranking, not detection, is the limiting stage, and the default configuration trails P2Rank on cryptic sites by 7.8 points with only the optional PLM ranker reaching parity.
Strengths
Metric-integrity analysis is exemplary
Section 3.2 quantifies recall gaming: ordering candidates by volume alone scores 77.7% under recall@0.30 — indistinguishable from the learned ranker — but 52.0% under Jaccard@0.25 vs 55.9%. The paper adopts Jaccard throughout and discloses that this analysis covers only 179 structures and the Jaccard clause.
Well-controlled MDpocket comparison
MDpocket receives the identical normal-mode ensemble, isolating clustering/ranking as the source of the +11.7-point gap (55.6% vs 43.9%, 95% CI +3.9 to +19.4), and is reported at its best of ten swept configurations, not its default.
Unusually candid negative results
Strong reproducibility engineering
Runtime claim is substantiated
Weaknesses
Test-fold hygiene is acknowledged but unquantified
Fragmented evaluation protocols invite invalid comparison
Missing head-to-head with PocketMiner
Selective statistical rigor
Reliance on an unpublished self-authored companion preprint
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
Lacuna (Sections 1–2) is a pip-installable tool that reframes cryptic-pocket discovery as ensemble bookkeeping rather than a simulation problem: generate cheap conformers (default anisotropic network model), detect pockets per conformer (fpocket-lineage alpha spheres), cluster detections across conformers into persistent sites, and rank sites with a 23-feature linear model fitted on within-structure pairs. The claims are routine usability (median 2.6 s/chain on one CPU core, Figure 4), pluggable sampling and detection, and a large ranking gain over the analytic rule it replaced (17.8% → 55.6% on identical candidates, Figure 3b). Critically, the paper pre-commits to honest scope: on already-open sites P2Rank wins (Section 3.2, COACH420 −6.9, CI −12.5 to −1.4); on cryptic sites the zero-dependency default trails P2Rank by 7.8 points (CI excluding zero), and only the optional PLM ranker reaches parity — 'parity, not a win,' in the authors' words.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 5 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is Lacuna and what problem does it solve?
Lacuna is an open-source Python tool that finds cryptic binding pockets — cavities that are absent or too small to detect in a protein's resting (apo) structure but open transiently during motion. Instead of scoring one static structure, it generates a cheap conformational ensemble, finds pockets in every conformer, clusters them into persistent sites, and ranks them; the default configuration runs in a median of 2.6 seconds per chain on a single CPU core.
How accurate is Lacuna at finding cryptic pockets?
On the CryptoBench test fold it recovers 55.6% of known cryptic sites in its top five predictions by default, 66.1% with an optional protein-language-model ranker, and 73.9% when pooling geometric and learned-surface detectors (measured under a separate five-conformer protocol). It slightly trails P2Rank on cryptic sites and clearly loses to P2Rank on already-open binding sites.
Is Lacuna free, and how do I install it?
Yes — it is MIT-licensed, installable with 'pip install lacuna-pockets', and archived on Zenodo (v1.2.0). The base install needs no GPU, no force-field setup, and no external neural-network checkpoints, and all benchmark scripts ship in the repository with automatic dataset downloads.
What are the main caveats about Lacuna's benchmarks?
The authors measured test-fold performance during development, so the headline numbers are held-out but not fully blind. The detector-fusion results use a different evaluation protocol than the headline numbers, no direct comparison with PocketMiner is provided, and P2Rank remains better on ordinary, already-open binding sites.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figures 1–4 are available only via their captions in the extracted text; figure content could not be independently verified.
- The companion detector-comparison study (ref 10) is an unpublished preprint and could not be checked.
- No per-structure results or raw benchmark output were inspectable; all judgments rest on the reported aggregates.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.