Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Search research articles

Related Experiment Videos

Beware of q2!

Alexander Golbraikh¹, Alexander Tropsha

¹Laboratory for Molecular Modeling, Division of Medicinal Chemistry and Natural Products, School of Pharmacy, University of North Carolina at Chapel Hill, 27599, USA.

Journal of Molecular Graphics & Modelling

|February 23, 2002

Summary

This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

Machine Learning Models with a Reject Option to Minimize Prediction Error: Application to Optical Properties of Dye Molecules.

Research square·2026

Same author

The ROBOKOP v1.0 knowledge graph system for exploring relationships between biomedical entities.

Scientific reports·2026

Same author

Binding Free Energies without Alchemy.

ArXiv·2026

Same author

CACHE Challenge #3: Targeting the Nsp3 Macrodomain of SARS-CoV-2.

Journal of chemical information and modeling·2026

Same author

Adaptive Machine Learning Framework for Optimizing the Affinity Purification of Adeno-Associated Viral Vectors.

Biotechnology and bioengineering·2026

Same author

Medicines, Diseases, Indications, and Contraindications (MeDIC): a foundational resource to support drug repurposing.

Nucleic acids research·2025

Same journal

Residue-level insights into SGLT2 inhibition and Nav1.5 selectivity of gliflozin derivatives: A molecular dynamics and pharmacophore-guided study.

Journal of molecular graphics & modelling·2026

Same journal

A benchmarking-informed structure-based virtual screening strategy targeting Lm-PTR1: Leveraging the Northern African natural products database.

Journal of molecular graphics & modelling·2026

Same journal

In Silico identification of natural and synthetic inhibitors targeting KRAS mutants (G12D, G12V, and G12C) and wild-type in pancreatic cancer.

Journal of molecular graphics & modelling·2026

Same journal

Structural evolution, mechanical and thermal stability of 7-40 mol% yttria-stabilized zirconia: First-principles investigation.

Journal of molecular graphics & modelling·2026

Same journal

Halide-encapsulated C<sub>24</sub> fullerenes as molecular redox hosts for alkali metals: A density functional theory study.

Journal of molecular graphics & modelling·2026

Same journal

Efficacy of Tinospora cordifolia bioactives as agonists of Smoothened (Smo) receptor to promote oligodendroglial lineage induction for remyelination-based therapy.

Journal of molecular graphics & modelling·2026

See all related articles

High leave-one-out cross-validation R2 (LOO q2) does not guarantee predictive ability in quantitative structure-activity relationship (QSAR) models. External validation is essential for reliable QSAR model development.

Area of Science:

Computational Chemistry
Cheminformatics
Drug Discovery

Background:

Quantitative structure-activity relationship (QSAR) models are vital for predicting drug efficacy and properties.
Model validation is critical to ensure the reliability and predictive power of QSAR studies.
Leave-one-out cross-validation (LOO q2) is a commonly used internal validation metric, often assumed to indicate high predictive ability.

Purpose of the Study:

To critically evaluate the reliability of leave-one-out cross-validation R2 (LOO q2) as a sole indicator of predictive performance in QSAR modeling.
To investigate the correlation between internal validation metrics (LOO q2) and external predictive ability using two-dimensional (2D) molecular descriptors and k-nearest neighbors (kNN) QSAR.
To establish robust criteria for assessing the true predictive power of QSAR models.

Related Experiment Videos

Main Methods:

Analysis of multiple datasets using two-dimensional (2D) molecular descriptors.
Application of the k-nearest neighbors (kNN) QSAR method.
Systematic comparison of LOO q2 values obtained from training sets against the predictive performance on independent test sets.

Main Results:

No significant correlation was found between high LOO q2 values and the predictive ability of the QSAR models on test sets across all analyzed datasets.
The study confirms that a high LOO q2 is a necessary but not sufficient condition for a QSAR model to possess strong predictive power.
This lack of correlation is suggested to be a general characteristic of QSAR models developed using LOO cross-validation.

Conclusions:

Relying solely on LOO q2 for QSAR model validation can be misleading, potentially overestimating a model's predictive capabilities.
External validation is indispensable for rigorously assessing and confirming the predictive performance of any QSAR model.
The study advocates for the implementation of a comprehensive set of criteria for evaluating QSAR model predictive ability, emphasizing external validation.