Skip to main content

Table 2 Comparison of performance of the SES models with different hyperparameters in the development dataset

From: Gene targeting in amyotrophic lateral sclerosis using causality-based feature selection and machine learning

 

p-value = 0.05

p-value = 0.03

p-value = 0.01

Cerebellum samples

Separation of ALS (both C9orf72-related familial ALS and sporadic ALS) from healthy samples

10 models

74.11%

81.48%

77.77%

5 models

66.66%

81.48%

85.18%

3 models

74.88%

77.77%

70.37%

Separation of C9orf72-related familial ALS from healthy samples

10 models

50.00%

33.33%

66.66%

5 models

72.22%

66.66%

83.33%

3 models

72.22%

72.22%

66.66%

Separation of sporadic ALS from healthy samples

10 models

33.33%

22.22%

38,88%

5 models

33.33%

22.22%

44.44%

3 models

33.33%

22.22%

33.33%

Frontal Cortex samples

Separation of ALS (both C9orf72-related familial ALS and sporadic ALS) from healthy samples

10 models

48.14%

51.85%

51.85%

5 models

51.85%

51.85%

55.55%

3 models

51.85%

48.14%

55.55%

Separation of C9orf72-related familial ALS from healthy samples

10 models

61.11%

72.22%

38.88%

5 models

61.11%

55.55%

44.44%

3 models

55.55%

50.00%

50.00%

Separation of sporadic ALS from healthy samples

10 models

61.11%

61.11%

61.11%

5 models

55.55%

61.11%

55.55%

3 models

55.55%

55.55%

50.00%

  1. Accuracy achieved by the SES regression models in the hold-out datasets for the separation of: (i) ALS (both C9orf72-related familial ALS and sporadic ALS) cerebellum from healthy cerebellum samples, (ii) C9orf72-related familial ALS cerebellum samples from healthy cerebellum samples, (iii) sporadic ALS cerebellum from healthy cerebellum samples, (iv) ALS (both C9orf72-related familial ALS and sporadic ALS) frontal cortex from healthy frontal cortex samples, (v)) C9orf72-related familial ALS frontal cortex samples from healthy frontal cortex samples, (vi) sporadic ALS frontal cortex from healthy frontal cortex samples. The referred accuracy is achieved due to taking into consideration: (a) the first 3 out of 10 SES models, (b) the first 5 out of 10 SES models and (c) all 10 SES models. The most accurate separation cases are highlighted in italics