Background Quantitative structure-activity relationship (QSAR) models are central to computer-aided drug discovery and predictive toxicology, but practical adoption is often impeded by ad-hoc tooling, inconsistent validation protocols, and poor reproducibility. Objective We introduce <Emphasis FontCategory="NonProportional">ProQSAR</Emphasis>, a modular, reproducible workbench that formalizes end-to-end QSAR development while permitting independent use of each component. Methods <Emphasis FontCategory="NonProportional">ProQSAR</Emphasis> composes interchangeable modules for standardization, feature generation, splitting (including scaffold- and cluster-aware splits), preprocessing, outlier handling, scaling, feature selection, model training and tuning, statistical comparison, conformal calibration, and applicability-domain assessment. The pipeline can run end-to-end to produce versioned artifact bundles (serialized models) and analyst-oriented reports suitable for deployment and audit. Results On representative <Emphasis FontCategory="NonProportional">MoleculeNet</Emphasis> benchmarks evaluated under Bemis–Murcko scaffold split, <Emphasis FontCategory="NonProportional">ProQSAR</Emphasis> attains state-of-the-art descriptor-based performance: the lowest mean RMSE across the regression suite (<Emphasis FontCategory="NonProportional">ESOL</Emphasis>, <Emphasis FontCategory="NonProportional">FreeSolv</Emphasis>, <Emphasis FontCategory="NonProportional">Lipophilicity</Emphasis>; mean RMSE <InlineEquation ID="IEq1"> <EquationSource Format="TEX">$0.658\pm 0.11$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>0.658</mn> <mo>±</mo> <mn>0.11</mn> </mrow> </math></EquationSource> </InlineEquation>), including a substantial improvement on <Emphasis FontCategory="NonProportional">FreeSolv</Emphasis> (RMSE <InlineEquation ID="IEq2"> <EquationSource Format="TEX">$0.494$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>0.494</mn> </mrow> </math></EquationSource> </InlineEquation> vs. <InlineEquation ID="IEq3"> <EquationSource Format="TEX">$0.731$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>0.731</mn> </mrow> </math></EquationSource> </InlineEquation> for a leading graph method). On quantum mechanical benchmarks, <Emphasis FontCategory="NonProportional">ProQSAR</Emphasis> demonstrated superior performance on the single-task dataset <Emphasis FontCategory="NonProportional">QM7</Emphasis> and maintained competitive results on the multi-task <Emphasis FontCategory="NonProportional">QM8</Emphasis> dataset. For classification, <Emphasis FontCategory="NonProportional">ProQSAR</Emphasis> achieves the top ROC–AUC on <Emphasis FontCategory="NonProportional">ClinTox</Emphasis> (91.4%) while remaining competitive across other benchmark (overall classification average <InlineEquation ID="IEq4"> <EquationSource Format="TEX">$70.4\pm 11.6$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>70.4</mn> <mo>±</mo> <mn>11.6</mn> </mrow> </math></EquationSource> </InlineEquation>). Crucially, all predictions are accompanied by cross-conformal prediction and explicit applicability-domain flags that identify out-of-distribution entries, enabling calibrated and decision support. Availability <Emphasis FontCategory="NonProportional">ProQSAR</Emphasis> is released on <Emphasis FontCategory="NonProportional">PyPI</Emphasis>, <Emphasis FontCategory="NonProportional">Conda</Emphasis>, and <Emphasis FontCategory="NonProportional">Docker Hub</Emphasis>; all releases embed full provenance (parameters, package versions, checksums) to ensure reproducibility. Scientific contribution <Emphasis FontCategory="NonProportional">ProQSAR</Emphasis> (i) enforces best-practice, group-aware validation together with formal statistical comparisons across models, (ii) integrates calibrated uncertainty quantification (cross-conformal prediction) and applicability-domain diagnostics for interpretable, risk-aware predictions, and (iii) exposes both a composable developer API and a one-click pipeline that generates deployment-ready artifacts and human-readable reports, demonstrated on representative benchmarks.

ProQSAR: A modular and reproducible framework for small-data QSAR modeling with fit-and-use models

期刊论文
出版模式：

开放获取
发表日期：

2026年4月22日

查看全文

Tuyet-Minh Phan,
Tieu-Long Phan,
Phuoc-Chung Van-Nguyen,
Lai Hoang Son Le,
Van-Thinh To,
Tuyen Ngoc Truong,
Daniel Merkle,
Peter F. Stadler

摘要

Background

Quantitative structure-activity relationship (QSAR) models are central to computer-aided drug discovery and predictive toxicology, but practical adoption is often impeded by ad-hoc tooling, inconsistent validation protocols, and poor reproducibility.

Objective

We introduce ProQSAR, a modular, reproducible workbench that formalizes end-to-end QSAR development while permitting independent use of each component.

Methods

ProQSAR composes interchangeable modules for standardization, feature generation, splitting (including scaffold- and cluster-aware splits), preprocessing, outlier handling, scaling, feature selection, model training and tuning, statistical comparison, conformal calibration, and applicability-domain assessment. The pipeline can run end-to-end to produce versioned artifact bundles (serialized models) and analyst-oriented reports suitable for deployment and audit.

Results

On representative MoleculeNet benchmarks evaluated under Bemis–Murcko scaffold split, ProQSAR attains state-of-the-art descriptor-based performance: the lowest mean RMSE across the regression suite (ESOL, FreeSolv, Lipophilicity; mean RMSE $0.658\pm 0.11$ $0.658 \pm 0.11$ ), including a substantial improvement on FreeSolv (RMSE $0.494$ $0.494$ vs. $0.731$ $0.731$ for a leading graph method). On quantum mechanical benchmarks, ProQSAR demonstrated superior performance on the single-task dataset QM7 and maintained competitive results on the multi-task QM8 dataset. For classification, ProQSAR achieves the top ROC–AUC on ClinTox (91.4%) while remaining competitive across other benchmark (overall classification average $70.4\pm 11.6$ $70.4 \pm 11.6$ ). Crucially, all predictions are accompanied by cross-conformal prediction and explicit applicability-domain flags that identify out-of-distribution entries, enabling calibrated and decision support.

Availability

ProQSAR is released on PyPI, Conda, and Docker Hub; all releases embed full provenance (parameters, package versions, checksums) to ensure reproducibility.

Scientific contribution

ProQSAR (i) enforces best-practice, group-aware validation together with formal statistical comparisons across models, (ii) integrates calibrated uncertainty quantification (cross-conformal prediction) and applicability-domain diagnostics for interpretable, risk-aware predictions, and (iii) exposes both a composable developer API and a one-click pipeline that generates deployment-ready artifacts and human-readable reports, demonstrated on representative benchmarks.