skip to main content
Idioma:

Be aware of error measures. Further studies on validation of predictive QSAR models

Roy, Kunal ; Das, Rudra Narayan ; Ambure, Pravin ; Aher, Rahul B.

Chemometrics and intelligent laboratory systems, 2016-03, Vol.152, p.18-33 [Periódico revisado por pares]

Elsevier B.V

Texto completo disponível

Citações Citado por
  • Título:
    Be aware of error measures. Further studies on validation of predictive QSAR models
  • Autor: Roy, Kunal ; Das, Rudra Narayan ; Ambure, Pravin ; Aher, Rahul B.
  • Assuntos: Dispersion ; Error measures ; MAE ; QSAR ; Validation
  • É parte de: Chemometrics and intelligent laboratory systems, 2016-03, Vol.152, p.18-33
  • Descrição: Validation is the most crucial concept for development and application of quantitative structure–activity relationship (QSAR) models. The validation process confirms the reliability of the developed QSAR models along with the acceptability of each step during model development such as assessing the quality of input data, dataset diversity, predictability on an external set, domain of applicability and mechanistic interpretability. External validation or validation using an independent test set is usually considered as the gold standard in evaluating the quality of predictions from a QSAR model. The external predictivity of QSAR models is commonly described by employing various validation metrics, which can be broadly categorized into two major classes, viz., R2 based metrics namely R2test, Q2(ext_F1), and Q2(ext_F2), and purely error based measures like predicted residual sum of squares (PRESS), root mean square error (RMSE), and mean absolute error (MAE). The problem associated with the error based measures is the absence of any well-defined threshold for determining the quality of predictions making the R2 based metrics more suitable for use due to easy comprehension. However, in this paper, we show the problems associated with the R2 based validation metrics commonly used in QSAR studies, since their values are highly dependent on the range of the response values of the test set compounds and their distribution pattern around the training/test set mean. We also propose a guideline for determining the quality of predictions based on MAE and its standard deviation computed from the test set predictions after omitting 5% high residual data points in order to obviate the influence of any rarely occurring high prediction errors that may significantly obscure the quality of predictions for the whole test set. In this manner, we try to evaluate the prediction performance of a model on most (95%) of the data points present in the external set. An online tool (XternalValidationPlus) for computing the suggested MAE based criteria (along with other conventional metrics) for external validation has been made available at http://dtclab.webs.com/software-tools and http://teqip.jdvu.ac.in/QSAR_Tools/. The MAE based criteria suggested here along with other commonly used validation metrics may be applied to evaluate predictive performance of QSAR models with a greater degree of confidence. •The external predictivity of QSAR models is commonly described by various validation metrics.•The values of R2 based metrics are dependent on the range and distribution pattern of the response values.•The error based measures do not have any well-defined threshold values.•A guideline is proposed here based on mean absolute error (MAE) and its standard deviation.•The MAE based criteria may be applied to evaluate predictive performance of QSAR models
  • Editor: Elsevier B.V
  • Idioma: Inglês
Voltar para lista de resultados
Ir para página anteriorAnterior Resultado  6 AvançarIr para próxima página

  • Realização: ABCD-USP USP   Logos de Redes Sociais: Freepik

Buscando em bases de dados remotas. Favor aguardar.