Table 1.

Results of classification of AD from CN

Top	Random forest				XGBoost				CNN
	Accuracy	STD(±)	AUC	STD(±)	Accuracy	STD(±)	AUC	STD(±)	Accuracy	STD(±)	AUC	STD(±)
100	66.46	7.79	0.7137	0.0576	70.24	2.80	0.7266	0.0281	68.29	2.87	0.7216	0.0603
200	67.18	3.88	0.7175	0.0386	67.99	1.52	0.7166	0.0245	69.52	5.08	0.7182	0.0494
300	66.26	3.80	0.7098	0.0377	68.20	3.32	0.7029	0.0272	70.64	2.20	0.7250	0.0585
400	67.58	4.67	0.7074	0.0428	69.42	3.43	0.7177	0.0234	67.99	4.65	0.7167	0.0412
500	67.59	7.79	0.7111	0.0457	71.05	2.56	0.7381	0.0325	71.56	6.58	0.7411	0.0614
1000	68.31	5.22	0.7178	0.0445	73.08	2.89	0.7407	0.0372	73.91	3.87	0.7741	0.0444
2000	68.70	3.13	0.7372	0.0424	72.48	2.61	0.7509	0.0365	73.29	2.77	0.7782	0.0409
3000	67.78	3.59	0.7282	0.0351	69.62	4.27	0.7376	0.0328	73.80	2.40	0.7862	0.0282
4000	68.19	4.69	0.7263	0.0469	71.15	4.07	0.7412	0.0368	75.02	3.17	0.8157	0.0261
5000	66.25	5.41	0.7105	0.0399	70.74	3.14	0.7330	0.0305	73.19	4.72	0.8003	0.0506
10 000	66.26	5.59	0.6919	0.0528	69.63	3.27	0.7248	0.0211	71.05	6.57	0.7083	0.1424

Notes: The table shows the number of top SNPs selected based on phenotype influence score for AD classification and the accuracy and AUC of 10-fold cross-validation. Our CNN-based approach yielded the highest accuracy and AUC of 75.02% and 0.8157, respectively, for 4000 SNPs. In all cases, our CNN models outperformed two traditional machine learning models, random forest and XGBoost