Chemistry Capstone · Computational Medicinal Chemistry
Comparing Candidate Antischistosomal Compounds by Computed Drug-Likeness
A reproducible cheminformatics comparison of eight schistosomiasis-relevant molecules — from the frontline drug praziquantel to repurposed antimalarials — using RDKit-computed descriptors and published drug-likeness rules.
Summary
This capstone uses computational medicinal-chemistry tools to compare candidate antischistosomal compounds by chemical structure, drug-likeness, and predicted properties. The goal is not to claim a new treatment, but to demonstrate how chemistry and computation can be used to evaluate drug candidates. The analysis is limited to computed properties and does not assess efficacy or clinical suitability.
Disclaimer
An educational computational analysis. No medical advice, clinical conclusion, or experimental validation of efficacy. All descriptors are software predictions, not measurements.
Abstract
Schistosomiasis is a neglected tropical disease that affected an estimated 250+ million people in recent years, and its global control depends almost entirely on a single drug, praziquantel. That reliance — with praziquantel’s known liabilities (poor water solubility, a bitter racemate, weak activity against juvenile worms) — motivates interest in alternative and repurposed chemotypes. Eight schistosomiasis-relevant molecules were selected and their physicochemical descriptors computed with RDKit, with each value cross-checked against the compound’s PubChem record. Only published drug-likeness criteria — Lipinski’s Rule of Five, Veber’s rules, QED, and the PAINS and Brenk structural-alert catalogs — were applied; no composite scoring system was defined.
All eight molecules satisfy Lipinski’s and Veber’s criteria with zero violations, so these rules do not distinguish them; the continuous QED index does, with praziquantel scoring highest (0.799). The structural-alert flags correspond to the artemisinin peroxide bridge and metrifonate’s organophosphate group, which are associated with the compounds’ antiparasitic activity, so an alert does not by itself indicate an unsuitable compound. A principal limitation is that praziquantel’s known formulation liabilities are not captured by any of the descriptors used here; the analysis is therefore framed in relation to the practical constraints of treatment.
Background
The disease
Schistosomiasis (bilharzia) is caused by parasitic flatworms of the genus Schistosoma, blood flukes that live in human blood vessels. Much of the harm results from the immune reaction to eggs trapped in tissue, causing chronic inflammation, organ damage, anaemia, and impaired growth in children. The parasite’s life cycle requires an intermediate freshwater-snail host, so control of the snail population can interrupt transmission. Global control relies on mass drug administration (MDA): periodic community-wide dosing, mostly of school-aged children.
Praziquantel’s treatment liabilities
Praziquantel is inexpensive, safe, and effective against adult worms of all major human species, but it has several liabilities that descriptor-based drug-likeness rules do not capture:
- Poor aqueous solubility. A BCS Class II drug (~0.4 mg/mL in water), so dissolution, not absorption, limits systemic exposure.
- A large, bitter racemate. Dosed as large 600 mg tablets at 40 mg/kg; only the (R)-enantiomer is active, while the inactive (S)-enantiomer contributes most of the bitterness, which complicates administration to young children.
- Reduced activity against juvenile worms. Immature worms are less susceptible, so recent infections can survive and mature after treatment.
- Reliance on a single drug. Near-total reliance on one compound under expanding MDA raises concern about reduced susceptibility and the need for alternatives.
These constraints — solubility, palatability, heat-stable formulation for tropical distribution, cost, and stage-specific activity — are the practical criteria relevant to treatment. They motivate the inclusion of older worm-active drugs, repurposed antimalarials active against juvenile stages, and the molluscicide used for snail control.
Research Question
Central question
How do selected antischistosomal and repurposed antiparasitic compounds compare in basic medicinal-chemistry properties and published drug-likeness metrics, and what can those computed comparisons indicate — and not indicate — about their suitability, in relation to the practical constraints of schistosomiasis treatment?
Methods
The analysis is reproducible and deterministic. Every descriptor is computed locally; no value is estimated manually.
- Compound selection & identifiers. Eight molecules chosen for their documented roles in schistosomiasis; each SMILES and PubChem CID retrieved from the PubChem PUG-REST API and re-verified by CID.
- Descriptor computation (RDKit 2026.03.3). MW, Crippen cLogP, TPSA, H-bond donors/acceptors, rotatable bonds, aromatic rings, fraction sp³, and QED.
- Cross-check against PubChem. Each value compared to PubChem for the same CID so that differences arising from descriptor definitions are made explicit.
- Published rules only. Lipinski (1997), Veber (2002), QED (Bickerton 2012), and the RDKit PAINS (Baell 2010) and Brenk (2008) catalogs. No composite score was defined.
- Comparison with measured data. Where public measured data exist (logP, solubility, melting point, BCS class), computed values are compared with them.
Relationship to my other work
This capstone is an independent, self-directed analysis of published chemical data. It is separate from my machine-learning schistosomiasis research project and does not report, restate, or depend on that project’s results.
Compound Selection
Eight molecules relevant to schistosomiasis treatment, grouped by role; the grouping is used as the colour encoding throughout the report.
- Frontline — Praziquantel. The current WHO standard of care for all human species.
- Historical — Oxamniquine, Metrifonate, Oltipraz. Once used, now largely retired (species-restricted, toxic, or superseded); useful benchmarks.
- Repurposing candidates — Artesunate, Artemether, Mefloquine. Antimalarials with reported antischistosomal activity, notably against the juvenile stages praziquantel misses.
- Transmission control — Niclosamide. The WHO molluscicide, applied to water to kill the snail host rather than taken by patients.
Chemical Property Analysis
The meaning of each descriptor, its relevance to oral drugs, and a comparison across the eight compounds.
For an oral drug, absorption depends on dissolution in the gastrointestinal tract and permeation across intestinal cell membranes. Several computed properties are used as indicators of these processes: molecular weight (an approximate size limit), cLogP (lipophilicity, the octanol–water partition), TPSA (topological polar surface area; higher values are associated with lower passive permeability, which declines markedly above ~140 Ų), hydrogen-bond donors and acceptors (which increase the desolvation energy required to cross a membrane), and rotatable bonds (a measure of molecular flexibility). All eight molecules fall within the ranges typical of small-molecule drugs.
Table 1 — Computed physicochemical descriptors (RDKit)
| Compound | Formula | MW | cLogP | TPSA | HBD | HBA | RotB | Ar.rings | Fsp³ | QED |
|---|---|---|---|---|---|---|---|---|---|---|
| Oltipraz | C8H6N2S3 | 226.3 | 3.30 | 25.8 | 0 | 5 | 1 | 2 | 0.125 | 0.551 |
| Metrifonate | C4H8Cl3O4P | 257.4 | 2.16 | 55.8 | 1 | 4 | 3 | 0 | 1.000 | 0.622 |
| Oxamniquine | C14H21N3O3 | 279.3 | 1.81 | 87.4 | 3 | 5 | 5 | 1 | 0.571 | 0.565 |
| Artemether | C16H26O5 | 298.4 | 2.84 | 46.1 | 0 | 5 | 1 | 0 | 1.000 | 0.697 |
| Praziquantel | C19H24N2O2 | 312.4 | 2.53 | 40.6 | 0 | 2 | 1 | 1 | 0.579 | 0.799 |
| Niclosamide | C13H8Cl2N2O4 | 327.1 | 3.86 | 92.5 | 2 | 4 | 3 | 2 | 0.000 | 0.661 |
| Mefloquine | C17H16F6N2O | 378.3 | 4.45 | 45.1 | 2 | 3 | 2 | 2 | 0.471 | 0.759 |
| Artesunate | C19H28O8 | 384.4 | 2.60 | 100.5 | 1 | 7 | 4 | 0 | 0.895 | 0.583 |
Drug-Likeness & ADMET Discussion
Published, citable rules only, with attention to what a rule violation or structural alert does and does not indicate.
All eight compounds pass Lipinski and Veber with zero violations, and none triggers a PAINS alert. Because all compounds satisfy these rules, the rules do not differentiate the set; the QED score and the structural-alert results provide the distinguishing information.
Table 2 — Published drug-likeness metrics (sorted by QED)
| Compound | QED | Lipinski viol. | Veber | PAINS | Brenk | Flagged groups (Brenk) |
|---|---|---|---|---|---|---|
| Praziquantel | 0.799 | 0 / 4 | pass | 0 | 0 | none |
| Mefloquine | 0.759 | 0 / 4 | pass | 0 | 0 | none |
| Artemether | 0.697 | 0 / 4 | pass | 0 | 1 | peroxide |
| Niclosamide | 0.661 | 0 / 4 | pass | 0 | 2 | Oxygen-nitrogen_single_bond; nitro_group |
| Metrifonate | 0.622 | 0 / 4 | pass | 0 | 2 | alkyl_halide; phosphor |
| Artesunate | 0.583 | 0 / 4 | pass | 0 | 1 | peroxide |
| Oxamniquine | 0.565 | 0 / 4 | pass | 0 | 2 | Oxygen-nitrogen_single_bond; nitro_group |
| Oltipraz | 0.551 | 0 / 4 | pass | 0 | 1 | Thiocarbonyl_group |
Interpretation of the Brenk alerts
The alerts flag the peroxide bridge of artesunate and artemether, the organophosphate group of metrifonate, the nitro groups of oxamniquine and niclosamide, and the thiocarbonyl of oltipraz. These substructures are treated as undesirable when assembling screening libraries. In several of these approved or historical drugs, however, the flagged group is essential to activity — for example, the endoperoxide bridge of the artemisinins. A structural alert therefore identifies a feature that warrants review, not evidence that a compound is unsuitable.
ADMET considerations
ADMET denotes absorption, distribution, metabolism, excretion, and toxicity. These descriptors relate only, and weakly, to absorption and distribution. Metabolism, excretion, and toxicity depend on specific enzyme interactions and reactive metabolites that cannot be inferred from MW, cLogP, or TPSA. No ADMET conclusions are drawn here; reliable ADMET assessment requires laboratory assays.
Results
The main findings are summarised below.
1. All compounds fall within typical small-molecule ranges, so the rule-based filters do not distinguish them. All 8/8 record zero Lipinski violations and 8/8 pass Veber, and no compound triggers a PAINS alert.
2. QED differentiates the compounds. Praziquantel (0.799) and mefloquine (0.759) score highest, while the atypical historical chemotypes — oltipraz (0.551) and oxamniquine (0.565) — score lowest. This is consistent with, but does not establish, differences in their development as oral drugs.
3. Recomputation reveals differences arising from descriptor definitions. Molecular weights agree to hundredths, but three systematic divergences from PubChem appear, each attributable to a definitional difference rather than an error:
- TPSA and sulfur. The N,O-only convention gives oltipraz 25.8 Ų, whereas PubChem’s sulfur-inclusive calculation gives 108 Ų.
- H-bond acceptors and fluorine. RDKit counts 3 acceptors for mefloquine (N and O only); PubChem counts 9 because it also counts the six fluorines.
- cLogP is method-dependent. For metrifonate, RDKit’s Crippen cLogP (2.16) is substantially higher than PubChem’s XLogP3 (0.5). The measured logP is approximately 0.5, which agrees with XLogP3 and indicates that the Crippen method overestimates lipophilicity for this organophosphate.
Table 3 — RDKit vs PubChem (definition divergences)
| Compound | MW rdkit/pubchem | cLogP rdkit | XLogP3 pubchem | TPSA rdkit (N,O) | TPSA pubchem | HBA rdkit/pubchem |
|---|---|---|---|---|---|---|
| Oltipraz | 226.35 / 226.30 | 3.30 | 1.10 | 25.8 | 108.0 | 5 / 5 |
| Metrifonate | 257.44 / 257.43 | 2.16 | 0.50 | 55.8 | 55.8 | 4 / 4 |
| Oxamniquine | 279.34 / 279.33 | 1.81 | 2.20 | 87.4 | 90.1 | 5 / 5 |
| Artemether | 298.38 / 298.37 | 2.84 | 3.10 | 46.1 | 46.2 | 5 / 5 |
| Praziquantel | 312.41 / 312.40 | 2.53 | 2.70 | 40.6 | 40.6 | 2 / 2 |
| Niclosamide | 327.12 / 327.12 | 3.86 | 4.00 | 92.5 | 95.2 | 4 / 4 |
| Mefloquine | 378.32 / 378.31 | 4.45 | 3.60 | 45.1 | 45.2 | 3 / 9 |
| Artesunate | 384.43 / 384.40 | 2.60 | 2.50 | 100.5 | 100.5 | 7 / 8 |
Table 4 — Computed vs measured logP, where measured data exist
| Compound | cLogP rdkit | XLogP3 pubchem | logP measured | Aq. solubility (meas.) | mp °C | BCS |
|---|---|---|---|---|---|---|
| Oltipraz | 3.30 | 1.10 | not found | water-insoluble; ~5 ug/mL (patent) | not found | not found |
| Metrifonate | 2.16 | 0.50 | 0.51 | 120-154 g/L (highly soluble, 25 C) | 82-84 | not found |
| Oxamniquine | 1.81 | 2.20 | 2.24 | ~0.3-0.82 mg/mL | 147-149 | not found |
| Artemether | 2.84 | 3.10 | 3.53 | practically insoluble in water | 86-88 | IV (reported) |
| Praziquantel | 2.53 | 2.70 | 2.5 | ~0.40 mg/mL (water, 25 C) | 136-138 | II |
| Niclosamide | 3.86 | 4.00 | not found | 1.6 mg/L (20 C); polymorph 0.6-13 ug/mL | 225-230 | II |
| Mefloquine | 4.45 | 3.60 | 3.9 | 1.806 mg/mL (free base) | 174-176 (free base) | not found |
| Artesunate | 2.60 | 2.50 | not found | very slightly soluble (~89 mg/L reported, unverified) | 131-135 | II (reported) |
Compared with measured values, RDKit’s cLogP is close for praziquantel and mefloquine, overestimates for metrifonate, and underestimates slightly for artemether. Computed logP therefore carries appreciable uncertainty, particularly for atypical structures. Two compounds (oltipraz and artesunate) have little or no publicly available measured data for comparison.
Limitations
The following limitations apply to this analysis.
- Predictions, not measurements. Descriptors carry model error, and logP and TPSA are convention-dependent, as shown above.
- No biological modelling. The analysis does not model target binding, mechanism, pharmacokinetics, resistance, or toxicity, and does not indicate antiparasitic activity.
- The heuristics apply to oral small molecules. They do not generalise to injectable or specially formulated drugs; niclosamide, for example, is applied to water rather than administered orally.
- Key treatment liabilities are not captured. Praziquantel’s solubility, taste, racemate composition, and reduced activity against juvenile worms are not reflected in any descriptor used here.
- No experimental validation. No assays were performed; laboratory and clinical testing are required to assess efficacy and safety.
Conclusion
Across the eight compounds, the rule-based filters (Lipinski and Veber) did not differentiate the set, whereas the QED index and the structural-alert results provided the distinguishing information. Praziquantel’s high computed drug-likeness coexists with documented treatment liabilities that these descriptors do not capture. The analysis demonstrates a reproducible cheminformatics workflow and its interpretation, and identifies the limits of a descriptor-based comparison.
Computational methods of this kind can prioritise candidates for further study, but they cannot establish efficacy or clinical suitability, which require experimental work.
References
- World Health Organization. Schistosomiasis — Fact sheet.
- U.S. CDC DPDx. Schistosomiasis (life cycle).
- Lipinski C.A. et al. (1997). Experimental and computational approaches to estimate solubility and permeability. Adv. Drug Deliv. Rev. 23:3–25. DOI:10.1016/S0169-409X(96)00423-1
- Veber D.F. et al. (2002). Molecular properties that influence oral bioavailability. J. Med. Chem. 45:2615–2623. DOI:10.1021/jm020017n
- Bickerton G.R. et al. (2012). Quantifying the chemical beauty of drugs (QED). Nat. Chem. 4:90–98. DOI:10.1038/nchem.1243
- Baell J.B., Holloway G.A. (2010). New substructure filters for removal of PAINS. J. Med. Chem. 53:2719–2740. DOI:10.1021/jm901137j
- Brenk R. et al. (2008). Lessons learnt from assembling screening libraries for neglected diseases. ChemMedChem 3:435–444. DOI:10.1002/cmdc.200700139
- Praziquantel Fifty Years On: A Comprehensive Overview of Its Solid State. Pharmaceutics (2024) 16:27. DOI:10.3390/pharmaceutics16010027
- N’Goran E.K. et al. (2023). Arpraziquantel orodispersible tablets in children, phase 3. Lancet Infect. Dis. 23:867–876. DOI:10.1016/S1473-3099(23)00048-8
- Summers S. et al. (2024). Praziquantel resistance in schistosomes. Front. Parasitol. 3:1471451. DOI:10.3389/fpara.2024.1471451
- PubChem — compound records by CID. RDKit — open-source cheminformatics. DrugBank — experimental physicochemical properties.
Code & Data Availability
Fully reproducible. The repository contains the compound definitions (SMILES + CIDs), the RDKit computation script, a Jupyter notebook that regenerates every table and figure, and the datasets.
build_dataset.py— RDKit descriptor computation used to generate the dataset.notebooks/antischistosomal_medchem_analysis.ipynb— reproducible analysis with outputs.data/compound_properties.csv— all computed descriptors, rules, and alerts.data/measured_properties.csv— cited measured values used for comparison.
Reproduce: pip install -r requirements.txt → run the notebook, or python build_dataset.py.