Fragment-based drug discovery (FBDD) starts by screening small, low-complexity molecules (molecular weight 100–300 Da, cLogP ≤ 3, ≤ 3 hydrogen bond donors) that bind with weak affinity but high ligand efficiency. The challenge: fragments bind at 100 µM–10 mM concentrations, requiring biophysical detection methods like SPR, NMR, or X-ray crystallography. Once a fragment hit is confirmed, the real work begins: converting a weak binder into a drug-like lead through growing, linking, or merging strategies. Computation plays a specific and well-defined role in this process.
Why fragment screening produces better starting points than HTS
High-throughput screening (HTS) tests large drug-like molecules (MW 400–600 Da) at high concentration. The hit rate is typically 0.01–0.5% — one in a thousand compounds tested shows activity. Fragment screening tests small molecules at higher concentrations against a smaller chemical space. The hit rate is much higher (1–10%) because fragments interact with binding sites through fewer but more energetically favorable contacts, and the chemical diversity of fragment space is more efficiently explored than drug-like space.
Ligand efficiency (LE = ΔG / heavy atom count) is the key metric for fragment hits. A fragment with Ki = 100 µM and 15 heavy atoms has LE = (−4 kcal/mol binding free energy) / 15 = 0.27 kcal/mol/atom. This is a reasonable starting point because there's room to grow the fragment toward higher affinity without sacrificing LE — adding 10 heavy atoms of well-designed elaboration can reach Ki = 10 nM while maintaining LE ≥ 0.25.
The risk: fragments tend to bind in the most energetically favorable sub-pocket, which isn't always the most druggable region for the full lead. FBDD campaigns sometimes start in binding sites that can't accommodate the elaborations needed to achieve drug-like ADMET properties.
Computational fragment docking
Before or alongside biophysical screening, computational docking of fragment libraries identifies candidate binding modes. Fragment docking differs from standard docking in important ways:
Library design. Fragment libraries used computationally (typically 5,000–50,000 fragments) are filtered for drug-like fragment properties: MW < 300, cLogP < 3, ≤ 3 H-bond donors, ≤ 3 H-bond acceptors, no reactive groups. Fragments with high molecular complexity (chiral centers, fused rings) are often excluded from initial screening because they sample conformational space more slowly.
Multiple binding sites. Fragment docking is routinely run across the full protein surface, not just the canonical active site. Allosteric sites, protein-protein interaction interfaces, and cryptic pockets are all accessible to small fragments. An exhaustive surface docking run identifies candidate pockets before committing to biophysical assay work.
Pose sampling depth. Fragments have fewer rotatable bonds than drug-like molecules, which means docking is faster per compound. But the small size of fragments means pose energies are less discriminating — a fragment fits many positions in a 4 Å sphere with similar docking scores. Clustering docking poses by binding mode, rather than selecting the single top score, is important for identifying the dominant binding hypothesis.
Fragment growing: extending into the binding site
Once a fragment hit is validated biophysically and a binding mode is proposed (ideally from crystallography), the growing phase begins. Growing attaches substituents to the fragment scaffold to extend into unexploited regions of the binding pocket, improving binding affinity.
Computational growing tools (Fragment Hopping, FTrees, FRED's fragment growing protocol) enumerate possible extensions from the fragment scaffold and score them against the protein structure. The output is a set of elaborated molecules for FEP validation before synthesis.
The FEP validation step is critical here. A computational growing tool can generate thousands of plausible elaborations. Without FEP ranking, the medicinal chemist must select synthesis candidates based on docking scores or visual inspection of binding poses — both imperfect. FEP on the top 50–100 growing candidates returns ΔΔG values that rank which elaborations are predicted to gain binding affinity from the extension.
Practical note on fragment growing: extensions that reach into solvent-exposed regions gain little binding affinity but add molecular weight. The most productive growing directions reach into enclosed hydrophobic pockets or form specific hydrogen bonds with residues that the parent fragment didn't contact. Visual inspection of the binding site to identify "unfilled" regions before designing growth vectors is worth the time.
Fragment linking: bridging two fragments in adjacent pockets
If two fragments are confirmed to bind in adjacent sub-pockets of the same binding site (usually established by crystallography of both fragments co-soaked), linking bridges the two fragments with a covalent tether. In principle, the free energy contribution of both fragments to binding is retained in the linked molecule, with a penalty for the linker's conformational entropy.
In practice, fragment linking is more difficult than it sounds. The distance and geometry between the fragments must be compatible with a short linker (typically 2–6 atoms). Longer linkers introduce flexible entropy that offsets the binding gain. Rigid linkers that constrain the optimal geometry (e.g., a phenyl ring or a short alkyl chain with minimal rotation) are preferred.
Computational linker design enumerates possible linker architectures that satisfy the distance constraints between the two fragment exit vectors. FEP on the top-ranked linker proposals validates which bridging chemistry is predicted to achieve additive binding affinity. The expectation is that a well-designed linked molecule will achieve affinity improvements of 2–4 orders of magnitude over either parent fragment (corresponding to 3–6 kcal/mol ΔΔG contribution from the linked elaboration).
Fragment merging
Fragment merging overlaps two fragments that share part of their binding pharmacophore, creating a single molecule that combines the binding contacts of both. Merging typically produces smaller, more synthetically tractable molecules than linking — the merged scaffold occupies the same binding volume as both fragments combined but without a linker.
Computational merging starts with identifying the overlap pharmacophore — the atoms in both fragments that make similar contacts with the protein. Tools like BREED (from Schrödinger) or Fragment Merging in ICM enumerate merged scaffolds. The resulting molecules often have better LE than the linked alternative because there's no conformational entropy penalty from a linker.
When to use FEP in FBDD workflows
FEP is most useful in FBDD at the elaboration stages, not initial fragment screening. At the initial screening stage (fragments typically binding at 100 µM–10 mM), the absolute binding affinities are too weak and uncertain for FEP to reliably rank-order. Once fragments have been validated biophysically and a binding mode is confirmed, FEP becomes reliable for predicting relative improvements from growing, linking, or merging strategies.
The practical guideline: use FEP when you have a confirmed fragment hit with a proposed binding mode (ideally crystallographic) and a defined elaboration strategy. FEP validates whether the proposed elaboration gains affinity from the new interactions. This substitutes synthesis-assay cycles at the hit-to-lead stage — the most expensive part of the FBDD workflow.
DrugSynq's fragment screening workflow accepts a fragment crystal structure, the identified binding mode, and a set of growing/linking proposals as SMILES. The pipeline returns FEP ΔΔG rankings within 48 hours. ADMET scoring is applied simultaneously — fragment elaborations often introduce ADMET liabilities as molecular weight and lipophilicity increase, and catching these computationally before synthesis is more efficient than discovering them in in vitro assays three weeks later.
