Four properties, one molecule

For a protein disclosed with its assay, its tool compound, and its counter-target, one new small molecule is drawn before any measurement on it and, in a single round of experiments, is potent, selective, soluble, and permeable.

A tight binder that cannot dissolve, cannot cross a membrane, or cannot ignore the protein next door does not become a drug. Projects learn those facts in sequence, and each fact starts a new design round. A method that hits all four on the first made molecule, for more than one target, would collapse that sequence into one experiment a standard laboratory can finish.

Pieces of the four properties have been predicted, or measured, separately. A blinded one-molecule round that hits all four has not.

The 2025 ASAP–Polaris–OpenADMET challenges asked participants to predict properties of compounds that had already been made in an antiviral program: potency at two coronavirus main proteases, and, separately, microsomal stability, kinetic solubility, log D, and MDR1–MDCK permeability (potency challenge (opens in a new tab), ADMET challenge (opens in a new tab)). The split was chronological and the test values were hidden. Predicting a number for an existing structure is a different task from drawing a new structure that then meets the numbers.

Atz and co-workers made two generated PPARγ ligands and measured several properties on them. Compound 1 had an EC50 of 1.5 ± 0.2 µM at PPARγ, 0.24 ± 0.05 µM at PPARδ, a calorimetric dissociation constant of 0.8 ± 0.1 µM, a log D of 1.5, and a PAMPA effective permeability of 3.9 × 10^-6 cm/s. Compound 2 had a selective EC50 of 2.3 ± 0.7 µM at PPARγ, a log D of 1.7, and a PAMPA permeability of 14 × 10^-6 cm/s. In a 50-target panel, both stayed under 50 percent effect at 10 µM except at PPARγ. The routes were six and five linear steps (Nature Communications, 2024, doi:10.1038/s41467-024-47613-w (opens in a new tab)). The paper reports log D and PAMPA. It does not report a kinetic-solubility pass against a threshold, and the campaign is two molecules for one receptor, chosen as the top of a generated list, with scoring models trained on known ligands.

Munson and co-workers synthesized 32 generated compounds aimed at MEK1 and mTOR together. Most had cell-growth IC50 values in the 1–10 µM range, and four of six retested compounds reduced both phosphorylation readouts by more than half at 1 µM. The authors state that the method, as published, does not provide molecules optimized for absorption, distribution, metabolism, excretion, and toxicity (Nature Communications, 2024, doi:10.1038/s41467-024-47120-y (opens in a new tab)).

Thomas and co-workers report an 8-of-9 binding hit rate, including three nanomolar ligands, for generated A2A compounds on a single receptor, with human filtering before synthesis (Nature Communications, 2025, doi:10.1038/s41467-025-60629-0 (opens in a new tab)). Selectivity against A2B was examined. Solubility and permeability were not the grade.

CACHE challenges #1 through #4 test many commercial compounds per team and confirm sparse hits, generally at micromolar concentrations. They are summarized in the companion problem on single-property design. They do not grade solubility and permeability on the same molecule that matched a tool compound.

An open doubt: PAMPA and a biochemical counter-target are not a full picture of permeability or selectivity in a person. They are the measurements a chemistry laboratory can finish in the same fortnight as the potency assay. A kinetic-solubility cutoff of 100 µM is a pass mark defined here, chosen so the result is an isolated number rather than a comparison with an insoluble tool compound. It is open to revision if a side-by-side standard becomes the field’s usual control.

The laboratory preregisters 4 protein targets. Each has a biochemical assay the grading laboratory already runs on a plate reader, by absorbance, fluorescence, or luminescence, a published tool compound, and a counter-target assay named in advance. The tool compound’s structure is not disclosed until the procedures are locked. The laboratory does disclose that its published half-maximal concentration in an assay of the same type lies between 10 nM and 1 µM. The four targets are disclosed together.

Within 48 hours, and before any new measurement on the designed compounds, the attempt submits one structure per target, a route of at most four linear steps from commercial materials, and a written procedure. One backup structure and route may be submitted at the same time and is opened only if the first procedure does not yield the compound. Nothing is redesigned after the first spectrum or the first assay plate.

The laboratory runs each procedure as written and isolates the product. Purity is at least 95 percent by quantitative proton NMR with a weighed internal calibrant. The compound passes the same Lipinski filter used in the companion design problem: at most one violation among more than 5 hydrogen-bond donors, more than 10 hydrogen-bond acceptors, molecular weight above 500, and calculated log P above 5. The laboratory checks that the structure is not a reported ligand of that target in ChEMBL or PubChem on the preregistration date.

Four measurements are then run on that isolated solid, in one round:

1. A dose–response of at least five concentrations, side by side with the tool compound and with vehicle.
2. The same format on the named counter-target.
3. Kinetic solubility at pH 7.4, by shake-flask and HPLC, or by a plate assay the laboratory has already validated, reported in micromolar.
4. Parallel artificial membrane permeability (PAMPA) at pH 7.4, reported as an effective permeability in cm/s.

The four targets may be synthesized and assayed in parallel. The whole round finishes within ten working days of the procedures being opened.

At least 3 of the 4 compounds pass all four of the following, on the isolated solid:

• half-maximal concentration within tenfold of the tool compound measured on the same plate
• at least tenfold weaker activity on the counter-target, with an inactive vehicle
• kinetic solubility of at least 100 µM at pH 7.4
• PAMPA effective permeability of at least 2 × 10^-6 cm/s

Three of four, the tenfold windows, 100 µM, 2 × 10^-6 cm/s, four linear steps, 95 percent purity, and the ten-day clock are pass marks defined here. The tenfold potency comparison uses the same plate, so assay drift does not decide the grade. The solubility and permeability numbers are pass marks, not records copied from a table.

Vehicle wells are inactive. For each target, a known inactive analog named at preregistration is inactive in the potency assay. The tool compound’s half-maximal concentration in the run lies within tenfold of its published value; otherwise the assay is repeated and not scored. Tenfold of the published value is a pass mark for assay validity.

On every solubility plate the laboratory runs a named high-solubility reference and a named low-solubility reference. The high reference must read at least tenfold higher than the low reference, or the plate is repeated. On every PAMPA plate the laboratory runs a named high-permeability reference and a named low-permeability reference. The high reference must exceed the low reference by at least fivefold, or the plate is repeated. The candidate is graded against the pass marks above, and the references only decide whether the plate is valid.

A procedure that does not produce the drawn compound fails that target. Purity below 95 percent fails that target even if the assays look favorable.

Any design method. Any reaction in the written route. One backup compound, submitted before any assay or spectrum of the test compounds. PAMPA lipids and solubility media that the laboratory names in the preregistration.

A compound that is already a reported ligand of the target. A second round of analogs after any result. A list graded on its best member. A peptide or an oligonucleotide. A Lipinski filter violation beyond the one allowed above. Quoting a published potency, solubility, or permeability in place of the side-by-side or same-week measurement. Substituting a cellular IC50 for the biochemical assay, or a calculated log P for the kinetic-solubility measurement. Changing the route after a failed step. A structure that differs from the tool compound, or from any reported ligand of that target, only by adding, removing, or replacing one methyl, fluoro, or hydroxy group. The tool structure is compared after the procedures are locked.

At least 3 of the 4 compounds meet the potency and selectivity bars, with purity and the drug-likeness filter satisfied, and miss solubility, permeability, or both. Three of four on those two endpoints is a pass mark for this lesser bar. Partial success is the companion single-property problem on a smaller set. It is not a solution of this one.

Lipinski, Lombardo, Dominy, and Feeney, Advanced Drug Delivery Reviews, 1997, 23, 3–25, doi:10.1016/S0169-409X(96)00423-1 (opens in a new tab). The “at most one violation” reading is a pass rule layered on that paper. Quantitative proton NMR purity follows Pauli and co-workers, Journal of Medicinal Chemistry, 2014, doi:10.1021/jm500734a (opens in a new tab).

The 100 µM solubility mark, the 2 × 10^-6 cm/s permeability mark, the tenfold windows, 3 of 4, and the four-step limit are pass marks defined here.

Cited experiments, each opened for this draft:

• Atz et al., Nature Communications, 2024, doi:10.1038/s41467-024-47613-w (opens in a new tab). Two PPARγ designs with the EC50, log D, and PAMPA values quoted above.
• Munson et al., Nature Communications, 2024, doi:10.1038/s41467-024-47120-y (opens in a new tab). Thirty-two dual MEK1/mTOR compounds; ADMET left for later.
• Thomas et al., Nature Communications, 2025, doi:10.1038/s41467-025-60629-0 (opens in a new tab). Eight of nine A2A binders on one target.
• Polaris challenges, 2025: antiviral potency (opens in a new tab) and antiviral ADMET (opens in a new tab). Blinded prediction for compounds already made.