GeinQSAR Studio
Home

Introduction

A Windows program for quantitative structure-activity relationships: from a table of structures and measured activities, or a query to ChEMBL or PubChem, to validated 2D and 3D models and the figures that report them.

GeinQSAR Studio takes a QSAR study through every step in one program. It reads a set of compounds with their activities, or builds such a set from ChEMBL or PubChem with a recorded curation, shows the compounds as a spreadsheet of structures, computes descriptors, designs the training and test sets, fits and validates 2D and 3D models, and exports the results as figures and reports of publication quality. The chemistry, the statistics, the 3D graphics and the interface are all part of the program; nothing else has to be installed.

The program as it opens. The welcome page in the centre offers the ways to start (a new or imported dataset, the Dataset Builder, a saved project, a new workflow), the six guided wizards and the recent files. The Project panel is on the left, the Inspector on the right, and the Log and Jobs buttons sit on the bar above the status line.
The program as it opens. The welcome page in the centre offers the ways to start (a new or imported dataset, the Dataset Builder, a saved project, a new workflow), the six guided wizards and the recent files. The Project panel is on the left, the Inspector on the right, and the Log and Jobs buttons sit on the bar above the status line.

What the program does

AreaWhat you get
DataImport of SD, MOL, MOL2, PDB and XYZ files, SMILES lists and CSV or tab-separated tables, with a column mapping you confirm before anything is read. A spreadsheet that draws every structure, filters by substructure and similarity, and holds formula columns. Standardisation, duplicate search, conversion of concentrations to a p-scale and six methods of training/test splitting.
Dataset BuilderA dataset prepared from a ChEMBL target or publication, a PubChem assay or your own file, curated in recorded steps that show what each step removed. A provenance file regenerates the dataset and compares it with the original.
DescriptorsPhysicochemical properties (Crippen logP, TPSA, hydrogen bonds, drug-likeness rules, QED, rings), a full 2D descriptor set, fingerprints, chirality-aware descriptors, and 3D families for compounds with conformers.
2D-QSARRegression and classification models built by a wizard that covers curation, descriptors, preprocessing, the split, descriptor selection, the model and its validation; dashboards with the statistics, the plots, the applicability domain and response randomisation; prediction of new compounds with a saved model.
3D-QSARConformers, alignment and two models: KEFRA, a lattice-free kernel method with a prediction uncertainty and an applicability domain, and atom-based cubes with PLS, with contour maps and atom cubes in an interactive 3D view.
PharmacophoresA common pharmacophore search over the actives of a series, with the hypotheses ranked and shown in 3D.
Library designLIBQORA builds the molecules a scaffold and a set of substituents imply, or single-point analogues of a hit, filters out what you do not want, predicts each one with your own models with a calibrated prediction interval and an applicability domain, and ranks them by objectives you name.
Chirality and activity cliffsStereoisomer pairs, activity cliffs measured by SALI, eudismic analysis, and a diagnosis of descriptors that cannot tell stereoisomers apart.
WorkflowsEvery wizard builds a workflow of nodes that you can open, change and run again. Ready-made templates start a complete analysis on your own data.
Figures and reportsCharts and 3D views exported as SVG or PNG at a stated size and resolution, and an report of a finished workflow, as PDF, Word and Excel, structured along the OECD principles.

What the results look like

The two pictures below come straight from the program. On the left is a 2D-QSAR model of dihydrofolate reductase inhibitors, on the right a 3D-QSAR model of the corticosteroid-binding globulin steroids that ship with the program as a sample.

Observed against predicted activity of a 2D-QSAR model. Training compounds are blue circles, their cross-validated predictions orange squares and the test compounds green triangles; the box gives N, R2, RMSE, Q2CV, Q2F1, Q2F3 and the test-set statistics, so the figure can stand alone in a paper.
Observed against predicted activity of a 2D-QSAR model. Training compounds are blue circles, their cross-validated predictions orange squares and the test compounds green triangles; the box gives N, R2, RMSE, Q2CV, Q2F1, Q2F3 and the test-set statistics, so the figure can stand alone in a paper.
A KEFRA contour map around aldosterone, with all six fields drawn at once in the colours the program uses throughout: green where more of the property raises the predicted affinity, red where it lowers it, each field in its own shade. The legend names the fields that are drawn and the bar at the bottom right gives the length scale. Drawing one field at a time, as the dashboard lets you, makes a single effect easier to read.
A KEFRA contour map around aldosterone, with all six fields drawn at once in the colours the program uses throughout: green where more of the property raises the predicted affinity, red where it lowers it, each field in its own shade. The legend names the fields that are drawn and the bar at the bottom right gives the length scale. Drawing one field at a time, as the dashboard lets you, makes a single effect easier to read.

How accurately the engines reproduce established reference implementations and published results, and under which conditions, is set out in Accuracy and validation.

Where to start

Installing and starting
What the package contains, what the computer needs, and the first start.
A tour of the window
Menus, ribbon, panels, documents, the status bar and the command palette.
Importing a dataset
Bringing your own structures and activities in, and what happens to bad rows.
Building a dataset from ChEMBL
Search, download and curation of public activity data in recorded steps.
The molecule spreadsheet
Structures, sets, filters, substructure and similarity search.
Molecular descriptors
The descriptor families and when each one is worth computing.
The 2D-QSAR wizard
From a prepared dataset to a validated regression model.
The 3D-QSAR wizard
Conformers, alignment, fields, PLS and contour maps.
Pharmacophore hypotheses
Common features of the actives of a series.
Chirality and activity cliffs
Stereoisomer pairs and the cliffs between them.
Designing a library
New molecules from a scaffold, predicted with your model and ranked.
The workflow canvas
The node graph behind every analysis.
Publication figures
Exporting charts and 3D views at a chosen size and resolution.

Two programs, one engine

The package holds two programs. GeinQSAR.exe is the desktop application described in this manual. gqsar-cli.exe is the command-line program: it runs a saved workflow without a window, computes descriptors, trains and applies models, converts structure files, builds a dataset from ChEMBL or PubChem, designs a library, and lists the node types and templates. Both use the same engine and read the same files, so a workflow designed in the window and saved as a .gqflow file gives the same result when it is run from a script, for example

gqsar-cli run --flow my-model.gqflow --report report.pdf

and a project saved by one machine opens on another with its workflows and dashboards. See The command line for every command.

Nothing is done behind your back. The program proposes and you decide: an import shows the columns it detected before reading anything, no conversion or standardisation is applied unless its box is ticked, and no rows are ever selected for you. Every wizard leaves a workflow that records exactly what it did.