GeinQSAR Studio
Home › Pharmacophore hypotheses

Pharmacophore hypotheses

A search for the arrangements of features that the actives of a series share and the inactives do not: the wizard, the ranked hypotheses, their geometry, the fit of each active, and the 3D view.

A pharmacophore hypothesis is a set of three to six feature points (acceptor, donor, hydrophobe, positive or negative ionisable group, aromatic ring) with fixed distances between them that the active molecules can all present at once. A hypothesis that the actives share and the inactives cannot present is a candidate for the features the binding site recognises. It is also an alignment rule: the actives superposed on their best hypothesis are an aligned set that can go into a 3D-QSAR study.

The search is GeinQSAR's formulation of the variant-tree approach of Dixon et al. (J. Comput.-Aided Mol. Des. 2006, 20, 647). Open it with Wizards > Pharmacophore hypotheses, or Pharmacophore in the Wizards group of the Workflow tab. Like every wizard it builds a workflow and runs it; the dashboard opens when the search finishes.

The wizard

1 Data

The Data page of the Pharmacophore hypothesis builder with the shipped acetylcholinesterase sample (data/flow/ache_pic50.csv, 429 compounds, activity pIC50).
The Data page of the Pharmacophore hypothesis builder with the shipped acetylcholinesterase sample (data/flow/ache_pic50.csv, 429 compounds, activity pIC50).

The same data page as every wizard (see The 2D-QSAR wizard). The activity must be on a logarithmic scale or be converted with Convert concentration to p-scale.

2 Actives and inactives

The series is divided at a threshold. The median of the activity column is offered as a suggestion, marked median, together with the resulting counts and a histogram in which the actives are blue and the inactives grey; set the value your assay calls active.

The threshold at the median, 6.19, with Active at or above: the page counts 11 actives and 9 inactives among the compounds read and draws the threshold as a line in the histogram.
The threshold at the median, 6.19, with Active at or above: the page counts 11 actives and 9 inactives among the compounds read and draws the threshold as a line in the histogram.
The same page for the 88 thrombin inhibitors: the median 6.69 divides them into 44 actives and 44 inactives.
The same page for the 88 thrombin inhibitors: the median 6.69 divides them into 44 actives and 44 inactives.
ControlDefaultMeaning
Active thresholdThe medianCompounds at or beyond it are the actives the hypothesis must explain; the other measured compounds are the inactives.
SenseActive at or aboveAt or above for a p-scale (pIC50, pKi); at or below for a concentration (IC50, Ki).
Penalise hypotheses inactives also matchOnA hypothesis the inactives match just as well explains nothing, so a weighted score of the inactives is subtracted from the survival score. Switch it off only when the inactives are not trustworthy (for example inactive for a reason unrelated to binding).

The page refuses a threshold that leaves fewer than two actives or no inactive compound.

3 Preparation

The Preparation page: standardised structures, 3D coordinates, conformers searched within a 10.0 kcal/mol window, and Search every conformer on. The maximum is set to 5 conformers here, a value for a quick run on a large sample.
The Preparation page: standardised structures, 3D coordinates, conformers searched within a 10.0 kcal/mol window, and Search every conformer on. The maximum is set to 5 conformers here, a value for a quick run on a large sample.
OptionDefaultMeaning and when to change it
Standardize structuresOnStrips salts and solvents, keeps the largest fragment and neutralises charges; the stereochemistry is kept.
Generate 3D coordinatesOnEmbeds one 3D structure per molecule; molecules that carry coordinates keep them unless the embedding options say otherwise.
Search conformersOnA flexible ligand needs a fair chance to present the pharmacophore, and the bioactive conformer is unknown.
Maximum conformers20Conformers kept per molecule, lowest energy first (1 to 200). The cost of the search grows with their number.
Energy window (kcal/mol)10.0Conformers more than this above the lowest are discarded.
Search every conformerOnEnumerates the sites of every stored conformer instead of the lowest-energy one only. This is the honest option, and it multiplies the cost of the enumeration.

4 Hypothesis search

The Hypothesis search page with hypotheses of three or four sites, a 1.0 Å distance bin, a 2.0 Å site separation, a 1.20 Å RMSD cutoff, half of the actives to match and ten hypotheses kept.
The Hypothesis search page with hypotheses of three or four sites, a 1.0 Å distance bin, a 2.0 Å site separation, a 1.20 Å RMSD cutoff, half of the actives to match and ten hypotheses kept.
OptionDefaultMeaning and when to change it
Minimum sites, Maximum sites3, 5Three points fix a rigid body. More sites are more specific but harder for every active to present at once, and the cost grows quickly with the count (3 to 10).
Distance bin (Å)1.0Tolerance of the inter-site distances: two arrangements whose distances fall into the same bins are one candidate. A larger bin merges more arrangements and finds more common hypotheses that are less precise.
Site separation (Å)2.0The smallest distance between two sites of one hypothesis; sites closer than this are not used together.
RMSD cutoff (Å)1.2An active matches a hypothesis when its sites superpose within this RMSD.
Actives to match0.50A hypothesis is kept when at least this fraction of the actives can present it. Raise it for a tight congeneric series, lower it when the actives belong to several chemotypes.
Hypotheses to keep20The best-ranked hypotheses that are listed and reported (1 to 200).

5 Output

The Output page with Report (PDF) and its title, and both further outputs ticked: the aligned actives as an SD file and the feature table as CSV, each with its file. Search starts the run.
The Output page with Report (PDF) and its title, and both further outputs ticked: the aligned actives as an SD file and the feature table as CSV, each with its file. Search starts the run.
OutputDefaultResult
Report (PDF)OnThe ranked hypotheses with their site geometry, how every active fits the best hypothesis, and the fit table.
Save the aligned activesOffThe actives as the search aligned them onto the best hypothesis, as a structure file. An SD file keeps their activities and data fields. Open it in the 3D-QSAR builder with the alignment As supplied to build a model on the alignment the hypothesis implies. An active that does not match is left out.
Save the feature tableOffThe pharmacophore features of every active, one per row: the molecule, the feature letter and type, the atoms that define it and its position. The dashboard then also shows the number of features of each type per active.

The dashboard

The dashboard has three tabs: Hypotheses, Geometry and 3D view. No hypothesis is selected when it opens; select one in the table or in the Hypothesis drop-down of the 3D view, and it becomes the subject of all three tabs.

The pharmacophore dashboard as it opens, on the Hypotheses tab: the ranked hypotheses with their variant, sites, survival scores and site score, Select a hypothesis under the table, and the features per active below. The inspector shows 10 hypotheses, 11 actives and 9 inactives and no selection.
The pharmacophore dashboard as it opens, on the Hypotheses tab: the ranked hypotheses with their variant, sites, survival scores and site score, Select a hypothesis under the table, and the features per active below. The inspector shows 10 hypotheses, 11 actives and 9 inactives and no selection.

The ranking

The terms of the score are reported separately, so that a ranking can always be explained.

ColumnDefinition
VariantThe feature letters of the hypothesis: A acceptor, D donor, H hydrophobe, N negative, P positive, R aromatic ring. A suffix (_1, _2) distinguishes hypotheses with the same letters and different geometry.
SitesThe number of feature points.
Site1 - RMSD / 1.2 Å of the matched sites, averaged over the matched actives.
VectorThe mean agreement of the matched direction vectors (cosine; absolute cosine for ring normals).
VolumeThe mean Gaussian shape Tanimoto of each aligned active with the reference active.
MatchesThe number of actives that present the hypothesis.
Survivalw(site) x Site + w(vector) x Vector + w(volume) x Volume + W(match)^(matches - 1).
Survival - inactiveSurvival minus a weighted survival evaluated on the inactives. A hypothesis the inactives present equally well loses rank.
Mean RMSDThe mean RMSD of the matched sites over the matched actives, in Å.

The weights are GeinQSAR defaults and are parameters of the search step of the workflow. Compare Survival with Survival - inactive: a hypothesis near the top by survival and low by the second column is common to the whole series, actives and inactives alike.

The ranking of a steroid search with ADH_1 selected: the cards below the table give its survival 3.9740, survival minus inactive 2.0547, site score 0.992, vector score 1.000, volume score 0.983, three actives matched and a mean RMSD of 0.010 Å.
The ranking of a steroid search with ADH_1 selected: the cards below the table give its survival 3.9740, survival minus inactive 2.0547, site score 0.992, vector score 1.000, volume score 0.983, three actives matched and a mean RMSD of 0.010 Å.
The ranking of the thrombin search: 20 hypotheses from 44 actives and 44 inactives, led by AHPRR_1 (an acceptor, a hydrophobe, a positive group and two aromatic rings) with a survival of 3.673, presented by 23 actives. AAHPR_1 and AAHHP_1 are presented by more actives (30) but score lower on the site term.
The ranking of the thrombin search: 20 hypotheses from 44 actives and 44 inactives, led by AHPRR_1 (an acceptor, a hydrophobe, a positive group and two aromatic rings) with a survival of 3.673, presented by 23 actives. AAHPR_1 and AAHHP_1 are presented by more actives (30) but score lower on the site term.

Geometry

The Geometry tab gives every inter-site distance and every angle of the selected hypothesis, with the score cards above. Below them, Aligned actives lists how each active fits the hypothesis: whether it was placed, the number and the letters of the sites it matches, the site RMSD, the fitness and a note.

The site geometry of ADH_1: the distances A1 - D2 4.17 Å, A1 - H3 3.23 Å and D2 - H3 4.02 Å.
The site geometry of ADH_1: the distances A1 - D2 4.17 Å, A1 - H3 3.23 Å and D2 - H3 4.02 Å.
The aligned actives of an acetylcholinesterase hypothesis: each active in view matches the four sites A1 A2 D3 R4 with an RMSD below 0.05 Å, and the line Alignment says the alignment was made by the search.
The aligned actives of an acetylcholinesterase hypothesis: each active in view matches the four sites A1 A2 D3 R4 with an RMSD below 0.05 Å, and the line Alignment says the alignment was made by the search.

Fitness of an aligned active is w(site) x (1 - RMSD / 1.2 Å) + w(vector) x vector + w(volume) x volume: the score a screening run would give that compound. For the best hypothesis the table and the molecules of the 3D view are the alignment the search itself made. The search publishes that alignment for its best hypothesis only, so for any other hypothesis the actives are superposed in the dashboard; the line Alignment under the table says which of the two is shown.

The Geometry tab of the thrombin hypothesis AHPRR_1: survival 3.6734, survival minus inactive 2.4158, site score 0.872, vector score 0.993, volume score 0.809, 23 actives matched with a mean RMSD of 0.153 Å. The aligned actives 01 to 05 match all five sites; 40 excluded volumes were derived.
The Geometry tab of the thrombin hypothesis AHPRR_1: survival 3.6734, survival minus inactive 2.4158, site score 0.872, vector score 0.993, volume score 0.809, 23 actives matched with a mean RMSD of 0.153 Å. The aligned actives 01 to 05 match all five sites; 40 excluded volumes were derived.

Features per active

Features per active, shown when the workflow wrote a feature table: the number of acceptors, donors, hydrophobes, negative and positive groups perceived on every active. A002 has eight acceptors and two donors.
Features per active, shown when the workflow wrote a feature table: the number of acceptors, donors, hydrophobes, negative and positive groups perceived on every active. A002 has eight acceptors and two donors.

A hypothesis can only use features that every matching active has, so this table explains why a compound with no donor never matches a hypothesis with a D in it.

Excluded volumes

An excluded volume marks a region that inactive compounds occupy and active compounds leave empty: space the receptor is presumed to fill. The search itself does not produce them; the dashboard derives them from the aligned molecules by a rule of its own, after the idea of Dixon et al. (2006):

  1. The candidates are the heavy atoms of the aligned inactives that are at least 2.2 Å from every heavy atom of every aligned active.
  2. The candidates are grouped by leader clustering with a radius of 1.6 Å, in a fixed order, so the result is deterministic.
  3. Each cluster becomes a sphere at its centroid, with a radius equal to the distance to the nearest active heavy atom minus 1.2 Å, limited to 0.9 to 1.6 Å. A sphere therefore never reaches into the actives.
  4. The clusters are ranked by the number of inactives that support them, and at most 40 are kept.

Without inactives, or when every inactive atom lies close to an active, there are none and the panel says so.

The 3D view

The 3D view of ADH_1 on the steroids: the acceptor (pink-red sphere with an arrow), the donor D2 (cyan with an arrow) and the hydrophobe H3 (green), each in its translucent tolerance sphere, the distances 3.23 and 4.17 Å and the angle 46.4°, with the three matching actives S11, S06 and S07 superposed. The Molecules section lists each active with its fitness.
The 3D view of ADH_1 on the steroids: the acceptor (pink-red sphere with an arrow), the donor D2 (cyan with an arrow) and the hydrophobe H3 (green), each in its translucent tolerance sphere, the distances 3.23 and 4.17 Å and the angle 46.4°, with the three matching actives S11, S06 and S07 superposed. The Molecules section lists each active with its fitness.
GlyphFeature
Pink-red sphere with an arrowHydrogen-bond acceptor and its direction
Cyan sphere with an arrowHydrogen-bond donor and its direction
Green sphereHydrophobe
Red sphereNegative ionic
Blue spherePositive ionic
Orange ringAromatic ring, drawn in the ring plane
Translucent sphere around a siteThe matching tolerance of the site
Teal sphereExcluded volume
ControlMeaning
HypothesisThe hypothesis shown, the same selection as the table.
ActivesThe actives aligned onto the hypothesis.
Tolerance spheres, Grey toleranceThe tolerance of each site as a translucent sphere; grey tolerance draws them in neutral grey, which leaves the colour to the feature glyphs and reads better in print.
Distances, AnglesLines and values of every inter-site distance; the first three inter-site angles.
Excluded volumesThe spheres described above; Excluded vol.: None when there are none.
MoleculesOne box per aligned active, with its fitness, to show or hide it.
View, AnnotationsAs in every 3D view; see The 3D workspace.
The 3D view before a hypothesis is chosen: the view says Select a hypothesis and the Hypothesis drop-down reads Select.
The 3D view before a hypothesis is chosen: the view says Select a hypothesis and the Hypothesis drop-down reads Select.
The thrombin hypothesis AHPRR_1 in the 3D view with the actives aligned onto it as sticks and the excluded volumes switched off; the site labels H2, R4 and P3 mark the features inside the overlay.
The thrombin hypothesis AHPRR_1 in the 3D view with the actives aligned onto it as sticks and the excluded volumes switched off; the site labels H2, R4 and P3 mark the features inside the overlay.

Figures

The best thrombin hypothesis AHPRR_1, matched by 23 actives, exported as a figure without the molecules: the acceptor A1 with its direction, the hydrophobe H2, the positive group P3 and the aromatic rings R4 and R5 in their tolerance spheres, with every inter-site distance as a dashed line and three of its angles (68.9°, 104.7° and 37.3°).
The best thrombin hypothesis AHPRR_1, matched by 23 actives, exported as a figure without the molecules: the acceptor A1 with its direction, the hydrophobe H2, the positive group P3 and the aromatic rings R4 and R5 in their tolerance spheres, with every inter-site distance as a dashed line and three of its angles (68.9°, 104.7° and 37.3°).
The same hypothesis with the actives the search superposed onto it, as sticks in one colour each. The aromatic and positive groups of the actives cluster at R4, R5 and P3.
The same hypothesis with the actives the search superposed onto it, as sticks in one colour each. The aromatic and positive groups of the actives cluster at R4, R5 and P3.

Both were made with the camera button of the 3D view (see Publication figures). A figure of the hypothesis alone, with its distances, is what a reader needs to use the hypothesis; the overlay of the actives shows how well it is supported.

A hypothesis is a model of what the actives have in common, not a binding mode. On a series with few features the pharmacophore alignment can leave the pose undetermined; on the steroids the features fix the pose for only 8 of 31 molecules, and the common-core alignment of the 3D-QSAR builder is the better choice there.