femiogundare/test

A Machine Learning Approach to Precise Renal Cancer Subtyping using RNA-seq Gene Expression Data

Objective

Cancer remains a significant global health challenge, with early detection and accurate classification being critical for effective treatment. In this repository, we develop an approach to detect the three most common subtypes of renal cancer — chromophobe renal cell carcinoma, clear cell renal cell carcinoma, and papillary renal cell carcinoma – by leveraging machine learning algorithms applied to RNA-seq gene expression data.

Results

The overall predictive performance of the machine learning methods based on ANOVA feature selection.

Model	Accuracy	Precision	Recall	Specificity	F1 Score	AUC
Logistic Regression	94.81	93.70	82.53	94.63	86.58	97.78
Naive Bayes	95.56	87.69	90.27	95.89	88.83	98.06
K-Nearest Neighbors	97.41	96.75	88.53	97.43	91.68	92.98
Support Vector Machine	97.04	96.74	85.63	96.69	89.64	98.23
Decision Tree	95.56	86.81	93.21	96.44	89.51	96.82
Random Forest Classifier	97.04	89.98	93.12	98.09	91.42	99.14
Extra Trees Classifier	96.30	87.88	92.20	97.38	89.84	98.49
Bagging Classifier	95.93	92.12	84.42	95.72	87.43	99.16

Results per class

The predictive performance of the machine learning methods per-class based on ANOVA feature selection.

Note: KICH stands for chromophobe renal cell carcinoma, KIRC stands for clear cell renal cell carcinoma, and KIRP stands for papillary renal cell carcinoma.

Classifier	Performance Metrics
	Class	Accuracy	Precision	Recall	Specificity	F1 Score
Logistic Regression	KICH	97.22	100.00	61.54	100.00	76.19
	KIRC	94.44	94.07	97.37	89.39	95.69
	KIRP	92.78	87.04	88.68	94.49	87.85
Naive Bayes	KICH	96.67	73.33	84.62	97.60	78.57
	KIRC	94.44	95.61	95.61	92.42	95.61
	KIRP	95.56	94.12	90.57	97.64	92.31
K-Nearest Neighbor	KICH	97.78	100.00	69.23	100.00	81.82
	KIRC	97.22	97.39	98.25	95.45	97.82
	KIRP	97.22	92.86	98.11	96.85	95.41
Support Vector Machine	KICH	97.22	100.00	61.54	100.00	76.19
	KIRC	96.67	95.76	99.12	92.42	97.41
	KIRP	97.22	94.44	96.23	97.64	95.33
Decision Tree Classifier	KICH	96.67	70.59	92.31	97.01	80.00
	KIRC	93.89	97.25	92.98	95.45	95.07
	KIRP	96.11	92.59	94.34	96.85	93.46
Random Forest Classifier	KICH	97.22	78.57	84.62	98.20	81.48
	KIRC	96.67	100.00	97.74	100.00	97.30
	KIRP	97.22	91.38	100.00	96.06	95.50
Extra Trees Classifier	KICH	96.67	73.33	84.62	97.60	78.57
	KIRC	95.56	99.07	93.86	98.48	96.40
	KIRP	96.67	91.23	98.11	96.06	94.55
Bagging Classifier	KICH	96.67	88.89	61.54	99.40	72.73
	KIRC	95.00	94.87	97.37	90.91	96.10
	KIRP	96.11	92.59	94.34	96.85	93.46

Fig. 1 | Receiver operating curve of Random Forest.

Fig. 2 | Receiver operating curve of Support Vector Machine.