Elsevier

Scientia Horticulturae

Volume 218, 14 April 2017, Pages 171-176
Scientia Horticulturae

Optimization of the number of evaluations for early blight disease in tomato accessions using artificial neural networks

https://doi.org/10.1016/j.scienta.2017.02.005Get rights and content

Highlights

  • ANNs were effective at predicting the area under Early Blight Disease progress curve.

  • Optimizing the assessment efficiency for disease severity reducing from six to two the number.

  • Evaluations to be carried out on the twelfth and eighteenth days after inoculation of this disease.

Summary

The efficacy of artificial neural networks (ANN) to solve complex problems can optimize evaluation processes for early blight disease on tomato plants, reducing required time and resources. The objective of the study was to verify the efficiency of ANN to predict the area under the disease progress curve (AUDPC) to reduce the number of assessments and establish the best time to evaluate early blight disease in tomato accessions. The severity of this disease was evaluated in one hundred and thirty-five tomato accessions from the Germplasm Vegetable Bank of the Federal University of Viçosa (BGH-UFV) in three experiments. The area under the disease progress curve (AUDPC) was calculated with data from six evaluations of the disease’s severity. Several ANN MLP types (Multi-Layer-Perceptron) were trained, taking into account AUDPC values for ​ desired output. Different numbers and assessment combinations for early blight disease severity were used as input. ANN’s were efficient at predicting AUDPC and reduced the number of evaluations from six to two. The twelfth and eighteenth days after pathogen inoculation are the best to evaluate the severity of early blight disease. Genotype by environment affects the efficiency in predicting the AUDPC. ANNs were efficient at predicting the area under the early blight disease progress curve (AUDPC) with fewer evaluations, and as such optimized assessment of this disease in tomato accessions.

Introduction

The tomato (Solanum lycopersicum L.), originally from South America, is one of the world’s most cultivated crops (Ranjan et al., 2012). It has a high yield, but is economically risky, mainly due to disease. Early Blight Disease (EBD), caused by the fungus Alternaria tomatophila Simmons, causes tomato crop losses (Brun et al., 2013).

A wide range of hosts, high variability, and a prolonged active stage all hamper the control of EBD (Singh et al., 2014). Relative humidity higher than 80% and moderate temperatures, of around 27 °C, favor the disease (Foolad et al., 2000). EBD is controlled primarily with fungicides. A few resistant tomato cultivars are also available (Ashrafi and Foolad, 2015). Chemical control is expensive and causes problems for humans and the environment (Hariprasad and Niranjana, 2009), however resistant cultivars can reduce this problem. This makes identification of sources of resistance in tomato accessions of germplasm banks highly valuable.

Data from six to eight evaluations of EBD severity processed at regular intervals, of- the area under the disease’s progress curve (AUDPC) (Mukherjee et al., 2010) can identify superior tomato genotypes for breeding programs (Grigolli et al., 2011, Foolad and Ashrafi, 2015, Rani et al., 2015). These assessments are labor-intensive and are usually carried out in the field using a large number of genotypes (Foolad et al., 2008, Kumar and Srivastava, 2013).

Artificial neural network (ANN) models were used to predict average regional wheat yields and production (Alvarez, 2009), environmental impact on strawberry production, (Khoshnevisan et al., 2013) and biological and environmental factors influencing single pea seed mass (Dacko et al., 2016). The efficiency of ANNs to model complex problems can predict AUDPC using fewer evaluations. This reduces labor and costs during selection of tomato accessions resistant to EBD. ANNs were also applied to predict genetic resource characteristics in plant breeding (Pandolfi et al., 2009, Bari et al., 2012, Emamgholizadeh et al., 2015). ANN computational models, which mimic the human brain in recognizing data patterns and regularities, represent an alternative as a universal substitute for complex functions (Gianola et al., 2011). They can have superior performance compared to conventional statistical models by being non-parametric, not requiring detailed information on the physical processes of the system studied, as well as tolerating data loss.

The objective of this study was to verify the efficiency of ANNs to predict the area under the progress disease curve (AUDPC) of Early Blight Disease (EBD) using fewer evaluations to establish the best time for its evaluation in tomato accessions.

Section snippets

Material and methods

The resistance of one hundred and thirty-five tomato accessions from the Germplasm Vegetable Bank of the Federal University of Viçosa (BGH-UFV) for Early Blight Disease (EBD) was evaluated in three experiments: the first from October 2009 to January 2010 with 33 accessions; the second from July to October 2010 with 51 accessions; and the third from September to December 2010 with 51 accessions (Laurindo et al., 2015). These accessions were randomly selected from approximately 840 accessions of

Results

The results from data with one evaluation showed less satisfactory results than that with two evaluations (Table 1). Average correlation of 0.19, −0.21, and 0.32 AUDPC between values calculated and predicted by the “Network 1”, “Network 2” and “Network 3”, respectively, was obtained. The best results were those from the fifth evaluation, with an average correlation between calculated and predicted AUDPC of 0.90, 0.91, and 0.91 for “Network 1”, “Network 2” and “Network 3”, respectively (Table 1).

Discussion

Artificial neural networks (ANN) simulate the biological neural system, mimicking the human brain’s learning processes to solve complex problems (Odabas et al., 2013). This technique has advantages of being non-parametric, tolerates data losses, and does not require detailed information on the system to be modeled (Silva et al., 2014) to solve complex problems of a linear or non-linear nature (Azevedo et al., 2015). Therefore, it may be useful to predict AUDPC with fewer evaluations.

The low

Conclusions

ANNs were effective at predicting the area under Early Blight Disease progress curve (AUDPC) in tomatoes, optimizing the assessment efficiency for disease severity and reducing from six to two the number of evaluations to be carried out on the twelfth and eighteenth days after inoculation of this disease.

Acknowledgements

To “Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq)” and “Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)”. Dr. Phillip Villani revised and corrected the English language used in this manuscript.

References (34)

  • J. Chelal et al.

    Modelling the interaction between early blight epidemics and host dynamics of tomato

    Trop Plant Pathol.

    (2015)
  • M.R. Foolad et al.

    Characterization of early blight resistance in a recombinant inbred line population of tomato: I. Heritability and trait correlations

    Adv. Stud. Biol.

    (2015)
  • M.R. Foolad et al.

    Comparison of field, green-house, and detached-leaflet evaluations of tomato germ plasm for early blight resistance

    Plant Dis.

    (2000)
  • M.R. Foolad et al.

    Genetics, genomics and breeding of late blight and early blight resistance in tomato

    Crit. Rev. Plant Sci.

    (2008)
  • D. Gianola et al.

    Predicting complex quantitative traits with Bayesian neural networks: a case study with Jersey cows and wheat

    BMC Genet.

    (2011)
  • J.F.J. Grigolli et al.

    Characterization of tomato accessions for resistance to early blight

    Crop Breed. Appl. Biotechnol.

    (2011)
  • P. Hariprasad et al.

    Isolation and characterization of phosphate solubilizing rhizobacteria to improve plant health of tomato

    Plant Soil

    (2009)
  • Cited by (0)

    View full text