STRC Research

Recipe — Fragment Optimization Linking Merging Growing

5 min read

Recipe — Fragment Optimization: Linking, Merging, Growing

P1 recipe — distilled from 2014-schneider-de-novo-molecular-design-book §5.3.4 (Durrant-Amaro, Fig. 5.1) and §6.5–6.7 (Mazanetz-Law-Whittaker). Covers the three optimization paths from a confirmed fragment hit to a lead-quality compound, with explicit selection criteria and failure modes for each path. Use this when h01 has a phase 4 → phase 4f confirmed fragment hit and needs to grow it into a compound capable of lead-class affinity (≤100 nM).

The three strategies (Fig. 5.1 — Schneider 2014 §5.3.4)

Strategy A — LINKING (Fig. 5.1a)

Two distinct fragments occupy adjacent (non-overlapping) subpockets. A linker is designed to join them into a single compound.

Use when:

  • Two ligand-efficient fragments crystallize in adjacent subpockets.
  • The geometry between fragments is well-defined (≤8 Å end-to-end).
  • A short, rigidifiable linker is feasible.

Avoid when:

  • Fragment binding modes are likely to shift on linking (Schneider 2014 §5.3.4 [64]: “linking frequently disturbs the binding modes of the original fragments”).
  • Required linker length is >8 Å (entropy penalty too large).

Failure mode: linker forms suboptimal interactions, neutralizing both anchor gains. Rigidifying the linker for entropic recovery is “often difficult” (Schneider 2014 §5.3.4 [78]).

Theoretical reward (when it works, §5.3.4 [8, 23, 28, 77]): “phenomenal” — sum of binding energies plus loss of one rotational+translational entropy ≈ +5 kcal/mol superadditivity. The factor Xa example in §1.5 shows −14 kJ/mol superadditivity for a single-bond linker (see Fragment Additivity Assumption and Superadditivity).

Strategy B — MERGING (Fig. 5.1b)

Two fragments overlap in the same binding subpocket. A composite single compound is built from their combined interacting moieties.

Use when:

  • Fragments share a common binding mode.
  • Substructures are partially overlapping but each contributes a distinct interaction (e.g., one fragment’s H-bond donor, another’s hydrophobe).

Avoid when:

  • Overlap is incomplete or geometrically strained.
  • Fragments have different scaffolds with no shared core.

Failure mode (Schneider 2014 §5.3.4): “binding fragments often fail to overlap in this ideal way” — applicable in <20% of FBDD hit pairs in the authors’ experience.

Strategy C — GROWING (Fig. 5.1c) — the default

Single fragment serves as anchor; derivatives are designed by attaching new groups in directions that exploit unused subpockets.

Use when:

  • One fragment is dominantly LE-efficient (LE ≥ 0.3 kcal/mol per HA).
  • Crystal/AF3 structure shows an unexploited subpocket adjacent to the anchor.
  • Synthetic vector for elaboration is accessible (carboxylic acid, amine, alcohol → coupling chemistry).

Avoid when:

  • The anchor fragment occupies the entire pocket (no unused real estate).
  • The anchor’s H-bond network is already saturated.

Failure mode: group-additivity assumption breaks down — see Fragment Additivity Assumption and Superadditivity. Mitigation: track Group Efficiency (GE; see Ligand Efficiency Metrics Catalog) at every elaboration step. If GE < 0.2 kcal/mol per added HA, revert.

Reward profile: linear, roughly 0.29 kcal/mol per added non-H atom (Schneider 2014 §6.4.1 [146, 222]). 38 HA at LE = 0.3 → ~10 nM affinity drug.

Decision matrix

Available evidenceRecommended strategy
Single dominant LE-efficient fragment + open subpocketGrowing
Two fragments in non-overlapping adjacent subpockets, ≤8 Å apartLinking
Two fragments overlapping in same subpocket with distinct interaction featuresMerging
Multiple fragments scattered with no clean geometric storyGrowing on best LE anchor
Crystallography unavailableGrowing on docked-pose anchor with explicit AF3/Boltz validation

Worked-example checklist for h01 phase 3c v4

Current state: 3-amino-benzofuran-2-COOH scaffold is the candidate anchor (per STRC Hypothesis Ranking h01 next-step). Pocket: E1659A subpocket (159 ų). Apply growing strategy:

  1. Anchor confirmation: verify LE of the 3-amino-benzofuran-2-COOH starting fragment via Vina + MM-GBSA (h01 phase 4f). Require LE ≥ 0.3 kcal/mol per HA. If lower, abandon and screen more fragments.
  2. Direction identification: inspect the WT and E1659A AF3 / Boltz pocket map. Highlight unfilled subpocket lobes (typically the K1141 hot-spot direction).
  3. Click-chemistry handle: the COOH and NH₂ on the anchor are both elaboration vectors. Build a virtual library of azide-alkyne click adducts (per Durrant-Amaro §5.3.4 / Fig. 5.2; Durrant 2012 PLoS Comput Biol — AutoClickChem in silico).
  4. Filter the elaboration library: apply the Recipe — Fragment Library Filtering Pipeline but with rule-of-five (drug-stage), MW ≤500 Da. Solubility ≥100 µM (relaxed from fragment ≥1 mM).
  5. Score by composite metric: sort by LE and LLE_AT (per Ligand Efficiency Metrics Catalog). Prefer compounds with Δ-LE ≥ 0 and Δ-LLE_AT ≥ 0 vs the parent fragment.
  6. Re-dock top 100 in WT and E1659A pockets (h01 phase 4 protocol).
  7. MM-GBSA on top 20 (h01 phase 4f).
  8. FEP point-mutation between top 5 analogue pairs to resolve sub-kcal differences (per Recipe — FEP Point-Mutation Algorithm) — but only after MM-GBSA agrees on the rank order. Do not invest in FEP for a series whose MM-GBSA scores span > 5 kcal/mol — those signals are within MM-GBSA error band.
  9. Track GE for every elaboration: GE < 0.2 kcal/mol per added HA = revert.

Citation pattern for h01 phase 3c v4 docstring

# Fragment-grow strategy (default for h01: single dominant anchor + unexploited E1659A subpocket).
# Method per Schneider 2014 §5.3.4 (Durrant & Amaro) and §6.5 (Mazanetz et al.) — fragment-growing
# is selected over linking/merging because (a) we have one LE-efficient anchor and (b) we have an
# unfilled subpocket. Group Efficiency tracked at every elaboration; revert if GE < 0.2 kcal/mol per HA
# (Schneider 2014 §6.4.1 [228]). Click-chemistry library constructed per AutoClickChem (Durrant 2012).

Connections

Graph View

Backlinks