Test one claim about Lacuna, the open-source cryptic-pocket tool: that it finds the KRAS switch-II pocket (SII-P) as a…
Test one claim about Lacuna, the open-source cryptic-pocket tool: that it finds the KRAS switch-II pocket (SII-P) as a site that opens with motion rather than one visible in the deposited coordinates. Design it as a paired comparison and run it with find_cryptic_pockets, chain A each time, with regions of interest "switch I" 30-38, "switch II" 60-76 and "SII-P groove" 95-102. 1) 4OBE, a GDP-bound KRAS with no inhibitor: here the SII-P should be closed, so if Lacuna is doing what it claims, the site should appear with high crypticity. 2) 6OIM, KRAS G12C covalently bound to sotorasib (AMG 510), which holds the SII-P open. Strip the ligand first (the tool reads protein atoms only) and note that you did. Here the same site should be near-zero crypticity, because it is already open in the coordinates. Then answer three things plainly: (a) does the SII-P show up in each run, and at which rank, crypticity, druggability and persistence; (b) do the lining residues match the residues that contact sotorasib in 6OIM (check the structure, do not recall them); (c) does the crypticity difference between the two structures go the direction the tool's claim predicts, or not. Say plainly if it does not. Keep the interpretation short and carry the limits: this is a normal-mode ensemble, not molecular dynamics, a pocket is geometry rather than evidence of binding, and two structures are an anecdote rather than a benchmark. Cite the Lacuna repository, its preprint, both PDB entries, and the sotorasib paper.
The paired test contradicts the tool's claim on this one structure pair: Lacuna scored the KRAS switch-II pocket (SII-P) as more cryptic in the sotorasib-open holo structure (6OIM) than in the ligand-free apo structure (4OBE) — the opposite direction from what the tool predicts. This is a single anecdotal pair scored by a normal-mode geometry heuristic, so it does not refute the method in general; it shows the crypticity difference did not run the predicted way here, and it points at why.
What I ran
Two find_cryptic_pockets runs, chain A each, NMA backend (21 conformers), ranked by crypticity, with ROIs switch I 30–38, switch II 60–76, SII-P groove 95–102.
- 4OBE — GDP-bound KRAS, no inhibitor, SII-P expected closed.
- 6OIM — KRAS G12C covalently bound to sotorasib (AMG 510), which wedges the SII-P open. I stripped the ligand first: the tool report shows only chain A protein was analysed (167 residues, no MOV/GDP heteroatoms in the pocket set), consistent with the note that it reads protein atoms only.
(a) Does the SII-P show up, and with what metrics?
Yes in both, but as different objects. I take "the SII-P" as the pocket with the highest SII-P-groove (95–102) coverage in each run.
| Structure | Best SII-P pocket | Rank | Crypticity | Druggability | Persistence | SII-P groove coverage | Switch II coverage |
|---|---|---|---|---|---|---|---|
| 4OBE (apo) | rank 5 | 5 of 5 | 0.000 | 0.716 | 0.333 | 0.375 | 0.471 |
| 6OIM (holo, ligand stripped) | rank 2 | 2 of 5 | 0.706 | 0.706 | 0.048 | 0.875 | 0.353 |
| 6OIM — larger overlapping pocket | rank 1 | 1 of 5 | 0.834 | 0.727 | 0.286 | 0.625 | 0.471 |
In 4OBE the SII-P surfaces only as the lowest-ranked pocket and is flagged not cryptic (crypticity 0.0 = already open in the deposited coordinates). In 6OIM the same region dominates the ranking (ranks 1–2) and is flagged cryptic at 0.706–0.834. Note the low persistence (0.048–0.286) — even where it scores cryptic it is not a stable feature of the ensemble.
(b) Do the lining residues match the sotorasib contacts?
I did not recall the contacts — I read them from the 6OIM header. SITE AC3 is defined there as the binding site for residue MOV A 303 (AMG 510, bound form); dropping GDP and waters leaves 16 protein contacts: V9, G10, C12, K16, P34, A59, G60, Q61, E62, E63, R68, M72, H95, Y96, Q99, V103 files.rcsb.org.
Intersecting those with the Lacuna 6OIM pocket linings (exact set intersection):
- Rank-1 pocket lining matches 15 of 16 sotorasib contacts — only H95 is missing.
- Rank-2 pocket (the one with the highest SII-P-groove coverage, 0.875) matches 7 of 16, and it is exactly the deep SII-P/α3 sub-shell: V9, R68, M72, H95, Y96, Q99, V103.
- The union of rank-1 + rank-2 covers all 16 contactsfiles.rcsb.org.
Method: I took the AC3 protein residue numbers and intersected them with each pocket's residue list — so the overlap is spatial correspondence, not evidence that a ligand binds. The pocket Lacuna calls the SII-P is genuinely the pocket sotorasib occupies. See the structure card "6OIM SII-P pocket (Lacuna rank 2) vs sotorasib contacts."
(c) Does the crypticity difference go the predicted direction?
No — it goes the opposite way, and plainly so. The claim predicts the apo SII-P should score high crypticity (a site that opens with motion) and the pre-opened holo site should score near-zero. Observed: apo 4OBE scored 0.0 and holo 6OIM scored 0.706–0.834. On this pair the tool ranks the already-widened pocket as the more cryptic one.
The mechanism is visible in the numbers rather than mysterious: crypticity here is scored relative to the input coordinates. 6OIM's backbone is frozen in the drug-expanded conformation, so with the ligand removed the tool sees a large empty volume that "opens" across the normal-mode ensemble and scores high. 4OBE's SII-P is already at its relaxed geometry, so the ensemble finds little additional opening and scores it 0.0. In other words, the heuristic is measuring opening relative to whichever conformer was crystallised, which is exactly why a holo-derived apo model can look more cryptic than a true apo model — a known hazard of scoring crypticity off a single ligand-biased structure.
Limits (carry these)
- This is a normal-mode ensemble, not molecular dynamics — a coarse CPU sampler; the tool itself notes a site that fails to open here could still open under MD/Boltz.
- A pocket is geometry, not binding evidence — the residue overlap with sotorasib is spatial, not thermodynamic.
- Two structures are an anecdote, not a benchmark. The 4OBE score of exactly 0.0 may also reflect the ROI/pocket-mapping choice or that this crystal form's SII-P is genuinely relaxed; one apo/holo pair cannot establish directional behavior of the crypticity metric. The honest statement is "on this pair, crypticity did not run in the predicted direction," not "the tool is wrong."
- Only chain A was analysed; interface pockets are invisible.
Sources
- Lacuna repository: https://github.com/mooreneural/lacuna
- Lacuna preprint: DOI 10.64898/2026.08.14.744956
- 4OBE (GDP-bound KRAS): PDB 4OBE, run
- 6OIM (KRAS G12C–AMG 510) coordinates and SITE AC3 contact list files.rcsb.org
- Sotorasib paper: Canon et al., Nature 575:217, 2019, PMID 31666701, DOI 10.1038/s41586-019-1694-1 (recorded from the 6OIM primary-citation header) files.rcsb.org
Research context, not experimental validation — these are computational pocket predictions on two crystal structures.