Peptide Screening — Computational Report

Composition-Preserving Scramble Screening: MM-GBSA Binding Energetics

Six top-ranked peptide hits against MMP1 and HYAL1, each scrambled into composition-preserving sequence variants and re-screened through the identical docking → MD → MM-GBSA pipeline used for the original hits, to test whether residue ordering can be improved without changing residue composition.

Pipeline status

Scramble screening — funnel counts

Counts of scrambled variants surviving each pipeline stage per parent, from initial generation through to the two stages (MD simulation, binding free energy) still in progress.

Peptide	Sequence	Target	Scrambled	3-Filter Pass	Folded	Hex OK	COM Filter Pass	HADDOCK3	Representative Structure	EM Verify	MD Simulation	Binding Free Energy
A596_338-343	FNQRQG	MMP-1	20	20	20	20	6	6	6	6	RESULTS BELOW	RESULTS BELOW
A123_18-23	TTARQL	MMP-1	20	19	19	19	10	10	10	10	RESULTS BELOW	RESULTS BELOW
B243_184-188	GQRQP	MMP-1	20	15	15	15	11	11	11	11	RESULTS BELOW	RESULTS BELOW
A1318orA1117	QPHQIA	HYAL1	20	19	19	19	1	1	1	1	RESULTS BELOW	RESULTS BELOW
B194_26-30	NGNNI	HYAL1	19	11	11	11	0	0	0	0	RESULTS BELOW	RESULTS BELOW
A67_191-195	HEAIR	HYAL1	20	20	20	20	2	2	2	2	RESULTS BELOW	RESULTS BELOW
Total			119	104	104	104	30	30	30	30

Column	Definition
3-Filter Pass	Toxicity, Skin Permeability, AntiAging
Folded	Getting 3D structure
Hex OK	Blind docking YES/NO
COM Filter Pass	Center of mass of peptide within 6Å of active site, at least 5 poses out of 100
HADDOCK3	Binding at active site
Representative Structure	Medoid of the biggest cluster after clustering HADDOCK poses, with at least 1 H-bond
EM Verify	Energy minimization

Binding Energetics

6 parents

Four of thirty scrambles outperform their original. Twenty-six do not. One — a single-position transposition against HYAL1 — outperforms its parent by nearly 25 kcal/mol.

Parent	Target	Sequence	Original ΔG	Scrambles	Best scramble	Outcome
A123_18-23	MMP1	TTARQL	−18.73	10	−21.53 (scr06)	beaten, modestly
A596_338-343	MMP1	FNQRQG	−26.46	6	−11.88 (scr03)	no scramble competitive
B243_184-188	MMP1	GQRQP	−17.05	11	−23.64 (scr01)	beaten, modestly
A67_191-195	HYAL1	HEAIR	−14.53	2	−39.25 (scr03) ★	beaten decisively
A1318orA1117	HYAL1	QPHQIA	−25.15	1	−6.35 (scr19)	n=1, worse
B194_26-30	HYAL1	NGNNI	−15.11	0	NO SCRAMBLED PEPTIDE QUALIFIED	no data filtered

A123_18-23

MMP1 · 10 scrambles

Original sequenceTTARQL

Original ΔG−18.73 ± 5.13 kcal/mol

Best scramble−21.53 (scr06)

Worst scramble+5.41 (scr12)

Variant	Sequence	ΔG (kcal/mol)	vs. original
scr06	LTTQAR	−21.53 ± 5.56	−2.8
scr01	QRTLAT	−16.48 ± 5.30	+2.3
scr05	ATQTLR	−15.38 ± 7.12	+3.4
scr18	RLTATQ	−14.84 ± 7.98	+3.9
scr11	TQATLR	−14.17 ± 5.58	+4.6
scr17	QATRTL	−12.18 ± 4.49	+6.6
scr03	LTARTQ	−11.00 ± 8.21	+7.7
scr13	LTAQRT	−6.20 ± 9.32	+12.5
scr10	LAQRTT	−2.82 ± 8.09	+15.9
scr12	TLARTQ	+5.41 ± 8.59	+24.1

Reading

One scramble (scr06) edges out the original by 2.8 kcal/mol — within the method's noise band, so at best a marginal, not decisive, improvement. The other nine range from flat to substantially worse, with scr12 losing binding altogether. The spread (−21.5 to +5.4 across ten orderings of the same six residues) shows this parent's binding is order-sensitive, but the native arrangement already sits near the top of what's achievable — scrambling mostly costs energy here rather than finding a better solution.

A596_338-343

MMP1 · 6 scrambles

Original sequenceFNQRQG

Original ΔG−26.46 ± 6.09 kcal/mol

Best scramble−11.88 (scr03)

Worst scramble+4.15 (scr01)

Variant	Sequence	ΔG (kcal/mol)	vs. original
scr03	FRQQGN	−11.88 ± 4.29	+14.6
scr12	NFQGQR	−11.65 ± 6.29	+14.8
scr07	QQNGFR	−8.58 ± 7.17	+17.9
scr04	RQQGNF	−8.27 ± 4.56	+18.2
scr15	FNQQGR	−7.30 ± 4.45	+19.2
scr01	RNQQFG	+4.15 ± 7.87	+30.6

Reading

The clearest negative result in the set. Every one of six scrambles underperforms the original, by margins of 14.6 to 30.6 kcal/mol — nothing here is close. This is the strongest evidence in the dataset that a native sequence's specific ordering, not just its residue composition, is load-bearing for binding: for this parent, the phenylalanine/asparagine/glutamine/arginine arrangement found by the original screen already appears close to a local optimum, and no random composition-preserving reshuffle recovers it.

B243_184-188

MMP1 · 11 scrambles

Original sequenceGQRQP

Original ΔG−17.05 ± 7.45 kcal/mol

Best scramble−23.64 (scr01)

Worst scramble+0.71 (scr07)

Variant	Sequence	ΔG (kcal/mol)	vs. original
scr01	QPGQR	−23.64 ± 7.71	−6.6
scr17	PRQQG	−19.27 ± 5.59	−2.2
scr05	QPQRG	−14.60 ± 7.20	+2.5
scr20	PQGQR	−13.23 ± 5.83	+3.8
scr03	RQGPQ	−13.01 ± 8.95	+4.0
scr04	RQQGP	−10.35 ± 6.32	+6.7
scr06	GPQRQ	−7.37 ± 14.65	+9.7 (high var.)
scr14	RPQQG	−3.96 ± 6.65	+13.1
scr18	QQPRG	−3.51 ± 6.35	+13.5
scr07	RQQPG	+0.71 ± 4.25	+17.8
scr13	GQQPR	+0.58 ± 2.00	+17.6

Reading

The largest scramble set, and the most graded result: two variants (scr01, scr17) beat the original by 2–7 kcal/mol, five sit within roughly 10 kcal/mol either side, and two lose binding entirely. Unlike A596_338-343, the native ordering here is not the ceiling — a genuine, modest improvement is available (scr01) — but nor is it arbitrary, since most reshuffles cost energy. This parent reads as "improvable but order-sensitive," a middle case between the other two MMP1 parents.

A67_191-195

HYAL1 · 2 scrambles

Original sequenceHEAIR

Original ΔG−14.53 ± 5.10 kcal/mol

Best scramble−39.25 (scr03) ★

Worst scramble+1.46 (scr12)

Variant	Sequence	ΔG (kcal/mol)	vs. original
scr03 ★	HEIAR	−39.25 ± 10.41	−24.7
scr12	RHAIE	+1.46 ± 11.93	+16.0

Reading

Only two orderings tested, and they land on opposite extremes: scr03 is the single strongest result across all thirty scrambles in this study; scr12 loses binding altogether. With n=2 this parent shows the highest ordering-sensitivity of the set — small positional changes swing ΔG by tens of kcal/mol in either direction. That volatility, combined with scr03 being the standout hit overall, made it worth a mechanistic follow-up rather than taking the number at face value.

Spotlight — why scr03 wins: a genuine binding-mode shift

scr03 is not a mutation — it's the same five residues as the original, with positions 3 and 4 (Ala, Ile) transposed: HEAIR → HEIAR. Receptor-ligand hydrogen-bond analysis (cpptraj, all 30 trajectory frames) shows this small reordering triggers a real change in how the peptide sits in the pocket, not a statistical fluctuation:

Original (HEAIR) — one dominant anchor

Arg422 (C-term) ↔ Asp110 66.7% occupancy
Arg422 ↔ Glu112 50.0%
Reciprocal (receptor→peptide) contacts: weak, ≤20% max

scr03 (HEIAR) — new anchor + reciprocal network

His418 (N-term, freed up) ↔ Asp110 93.3% — new
Arg422 ↔ Asp273 (different partner) 40–43%
Arg115 ↔ peptide C-terminus 90.0% — new
Thr124 ↔ Arg422 66.7% — new

The transposition frees His418 to reach Asp110 directly (previously occupied by Arg422), redirects Arg422 onto a different acidic residue, and opens an almost entirely new reciprocal receptor→peptide network anchored on Arg115/Thr124. The energy decomposition agrees: the ΔΔG is driven mainly by a large electrostatic gain (ΔEEL −95.6 kcal/mol) partly offset by a larger desolvation penalty (ΔEGB +86.8), plus a real van der Waals packing improvement (ΔVDW −13.0) — consistent with a genuinely different, more extensively contacted pose rather than noise.

A1318orA1117

HYAL1 · 1 scramble

Original sequenceQPHQIA

Original ΔG−25.15 ± 8.90 kcal/mol

Only scramble−6.35 (scr19)

Variant	Sequence	ΔG (kcal/mol)	vs. original
scr19	PQHQIA	−6.35 ± 5.85	+18.8

Reading

Only one scramble survived the pipeline for this parent, and it is substantially worse (+18.8 kcal/mol). With a sample of one, this cannot support a general claim about ordering sensitivity — it simply shows this particular reordering doesn't help. More scrambles would be needed before drawing any real conclusion for this parent.

B194_26-30

HYAL1 · 0 scrambles

Original sequenceNGNNI

Original ΔG−15.11 ± 8.06 kcal/mol

Scrambles generated19

Reached MM-GBSA0

No scramble comparison exists for this parent — not because binding is poor, but because the pipeline dropped every variant before energetics were ever computed. All 19 generated scrambles were hex-docked; 11 survived that stage; zero passed the active-site center-of-mass filter required to proceed to HADDOCK3 refinement, MD, and MM-GBSA. This is a data gap in the screening geometry, not a negative result — worth revisiting the filter threshold for this parent specifically before concluding anything about its scrambled variants.

Conclusion

Native ordering usually wins. 26 of 30 scrambles are weaker than their parent, several non-binding outright. Composition-preserving reshuffling is far more likely to cost binding energy than to find it — consistent with these six parents already being rationally-selected hits from prior screening, not raw random sequences.
But when it wins, it can win big. A67_191-195_scr03 beats its parent by ~25 kcal/mol via a single positional swap (Ala↔Ile) — confirmed structurally, not just numerically, via a genuine shift to a new, more extensively hydrogen-bonded binding pose. This is the priority candidate to carry forward.
HYAL1 parents show the widest swings — the single best result (scr03) and one of the worst (scr12, same parent) both come from HYAL1-targeting peptides, while MMP1 parents cluster more moderately. Sample sizes here are small (2, 1, and 0 scrambles for the three HYAL1 parents), so this reads as a signal worth testing further, not yet a settled trend.
One parent has no answer at all. B194_26-30 lost all 19 of its scrambles to the active-site filter before any energetics were computed — a pipeline gap, not a finding. Revisiting that filter's threshold for this parent is a concrete, low-cost next step.

Method notes & caveats

Protocol verified identical to the pre-scramble original screen — same config file (igb=5, saltcon=0.15), same 30-frame / 5–10 ns sampling window, same per-frame topology construction — so scrambled and original ΔG values are directly comparable, not just directionally similar.
Each pose has exactly one 10 ns MD trajectory; no independent replicate runs. Standard deviations of 5–15 kcal/mol are typical for 30-frame MM-GBSA and mean differences smaller than that shouldn't be over-read as meaningfully distinct.
MM-GBSA (implicit solvent, no entropy term) is a fast ranking method, not a substitute for FEP/TI or experimental binding data if A67_191-195_scr03 moves toward further validation — a confirmatory replicate run is recommended before treating −39.25 kcal/mol as final.
One result (A67_191-195_scr12) was initially computed on an incomplete trajectory (3 of 30 frames, a pipeline race condition) and was caught, discarded, and correctly recomputed once its MD run finished — the value shown here is the corrected, full 30-frame result.