PFASGroups GUI

Purpose

Load a molecular dataset and classify each molecule against the built-in PFAS group definitions. Matched groups are highlighted as colour-coded fragments in the structure view (Tab 2).

Data sources

• File – CSV, Excel (.xlsx/.xls), or SQLite (.db/.sqlite). Browse for the file, choose a sheet (Excel) and the SMILES / name columns.
• Database – SQLite, MariaDB, or PostgreSQL. Fill in the connection form (host, port, database name, user, password, table name).
• Manual – Paste SMILES directly. One per line, or comma-separated with optional names (Name: SMILES).

Options

• Halogens – Which halogens to include in the analysis. Start with F only for classical PFAS; broaden to include Cl/Br/I for expanded screening.
• Saturation filter – Perf. (CF₂/CF₃) restricts to perfluorinated chains; All fluorinated includes partially fluorinated structures.
• Compute metrics – Calculate graph-theoretic component metrics (needed for Tabs 4–6).
• Check definitions – Additionally test each molecule against the PFAS definitions ticked below the option.

Purpose

Browse the classified molecules as interactive compound cards. Each card shows:

• The 2-D structure with matched PFAS groups highlighted in group-specific colours.
• A list of matched group names and component counts.
• PFAS definition badges (pass/fail) if definition checking was enabled.

Controls

• Search – filter cards by molecule name.
• Show All / Hide All – toggle structure visibility for all cards.
• Export CSV – save a summary table (name, SMILES, matched groups, definition results) to a CSV file.

Sub-tab A – PFAS Definitions

Enter a SMILES string and test it against:

• Built-in definitions (default, when the custom JSON field is left blank).
• Custom PFASDefinition JSON – paste an array of definition objects. Each must have id, name, smarts (list of SMARTS strings), and optionally fluorineRatio and description.

Sub-tab B – Custom Group (SMARTS)

Define a custom HalogenGroup and test a molecule against it. Fields:

• componentSmarts – name of the underlying fluorine component pattern (e.g. Perfluoroalkyl).
• SMARTS dict (JSON) – mapping of SMARTS pattern → required minimum count, e.g. {"[CX3](=O)[OX2H1]": 1}.
• Constraints (JSON) – element-count constraints, e.g. {"gte": {"F": 4}}.

Diagnostics are printed as JSON in the right panel, showing which stage passed or failed (component detection, SMARTS matching, constraint checking).

Purpose

Rank the classified molecules to identify the most structurally diverse or representative candidates.

Modes

• Intrinsic: total component size – sum of all matched fragment sizes.
• Intrinsic: max component size – largest single matched fragment.
• Intrinsic: total + max – weighted combination; use α and β to balance the two terms.
• Reference-based KL divergence – divergence from a supplied reference dataset. Supply a CSV/Excel/SQLite file as the reference.

Output

A sortable ranking table and a horizontal bar chart showing the top-30 priority scores. Higher scores indicate higher priority (more structural novelty or complexity relative to the reference).

Purpose

Visualise the PFASGroups fingerprint space as an interactive 2-D scatter plot.

Methods

• UMAP – fast, topology-preserving. Adjust n_neighbours (local structure) and min_dist (cluster compactness).
• PCA – linear; no hyper-parameters.
• t-SNE – non-linear; adjust perplexity.

Fingerprint presets

Select a FINGERPRINT_PRESETS key. best (binary + effective_graph_resistance) gives the best inter-group discrimination according to the MQG benchmark, but any preset can be used.

Colour by

By default, points are coloured by the dominant PFAS group. If a label column was loaded with the data, select it here to colour by an arbitrary property.

Purpose

Compare fingerprint descriptors for binary property prediction using HistGradientBoostingClassifier with repeated stratified k-fold CV.

Target column

Choose a column from your loaded data that contains binary labels (0/1 or True/False). Common examples: bioactive, toxic, active.

Fingerprint sets

• PFASGroups presets – any combination of the benchmark-validated fingerprint presets (binary, best, best_2, …).
• Morgan (RDKit) – radius-2 Morgan fingerprint, 512 bits. Useful baseline.
• ToxPrint (729 bits) – requires pyCSRML to be installed.
• TxP_PFAS (129 bits) – PFAS-specific ToxPrint subset; requires pyCSRML.
• Custom TSV/CSV – rows = molecules (same order as input), columns = bits. No header row.

Bayesian correlated t-test

All fingerprint-set pairs are compared using the Bayesian correlated t-test (Benavoli et al. 2017) with ROPE = 0.01 ROC-AUC. The table reports:

• P(A>B) – probability that set A is meaningfully better than B.
• P(ROPE) – probability that the two sets are practically equivalent.
• P(B>A) – probability that set B is meaningfully better than A.

• Ctrl+Q – quit the application.
• Ctrl+W – close (same as Ctrl+Q on Windows).

Install the GUI dependencies into your environment:

pip install "PFASGroups[gui]"

Or from the repository:

pip install -e ".[gui]"

Then launch with:

pfasgroups-gui

• Benavoli A et al. (2017) Time for a Change: a Tutorial for Comparing Multiple Classifiers Through Bayesian Analysis. JMLR 18(1):2653–2688.
• Nadeau C & Bengio Y (2003) Inference for the Generalization Error. Machine Learning 52(3):239–281.