Skip to main content

Table 3 The AUC for various methods averaged over cancer types

From: Multiple instance neural networks based on sparse attention for cancer detection using T-cell receptor sequences

Method

ADeep

BoW

CCE

CkNN

EMD-SVM

Balance

66.44 (8.18)

68.74 (1.60)

69.89 (1.17)

51.97 (1.85)

72.67 (1.23)

Imbalance

70.48 (15.33)

64.28 (4.90)

61.91 (3.82)

52.33 (3.65)

70.19 (2.04)

Method

EMDD

mi-Net

mi-SVM

MI-SVM

miGraph

Balance

56.98 (3.81)

67.66 (9.03)

67.08 (1.48)

67.87 (1.58)

63.88 (1.49)

Imbalance

62.67 (4.67)

63.79 (17.66)

64.69 (3.50)

66.55 (2.94)

59.50 (2.62)

Method

MILBoost

MILES

MInD

MINet

MINN-SA

Balance

50.83 (2.31)

68.94 (1.40)

67.03 (1.66)

58.07 (14.25)

73.88 (8.82)

Imbalance

59.89 (2.45)

66.05 (3.90)

65.94 (2.85)

35.28 (16.10)

79.20 (17.18)

Method

NSK-SVM

SI-kNN

SI-SVM

Average

Balance

70.00 (1.31)

66.50 (1.48)

66.72 (1.61)

64.78 (12.39)

Imbalance

63.43 (3.12)

57.63 (5.16)

65.05 (2.94)

61.74 (16.80)

  1. The number in parenthesis is the average standard deviation of AUC from 10-fold CV across cancer types. The last method (“Average”) shows the mean and the standard deviation of AUC values from all methods except MINN-SA. Numbers in boldface are the best 3 methods in each of balanced and imbalanced cases