Towards a potential pan-cancer prognostic signature for gene expression based on probesets and ensemble machine learning

Chicco, Davide; Alameer, Abbas; Rahmati, Sara; Jurman, Giuseppe

doi:10.1186/s13040-022-00312-y

Table 2 Results obtained by our pan-cancer signature on 57 gene expression datasets

From: Towards a potential pan-cancer prognostic signature for gene expression based on probesets and ensemble machine learning

	dataset name	cancer type	MCC	F₁ score	accuracy	TPR	TNR	PPV	NPV	PR AUC	ROC AUC
1	dataHeaton2011	adrenocortical cancer	+0.082	0.401	0.404	0.852	0.245	0.318	0.813	0.617	0.631
2	dataReister2012	bladder cancer	+0.061	0.557	0.471	0.844	0.204	0.428	0.669	0.501	0.541
3	dataHatzis2009	breast cancer	+0.252	0.414	0.723	0.466	0.801	0.406	0.839	0.395	0.697
4	dataYenamandra2015	breast cancer	+0.219	0.509	0.568	0.738	0.490	0.417	0.798	0.511	0.648
5	dataJezequel2015	breast cancer	+0.189	0.799	0.703	0.854	0.308	0.759	0.479	0.835	0.654
6	dataSchmidt2008	breast cancer	+0.162	0.347	0.691	0.381	0.784	0.353	0.811	0.352	0.643
7	dataMiller2013	breast cancer	+0.151	0.277	0.634	0.561	0.650	0.207	0.908	0.267	0.640
8	dataSinn2019	breast cancer	+0.148	0.898	0.822	0.923	0.204	0.876	0.304	0.903	0.621
9	dataDesmedt2007	breast cancer	+0.129	0.424	0.520	0.681	0.462	0.317	0.802	0.360	0.597
10	dataKarn2011	breast cancer	+0.104	0.313	0.449	0.227	0.871	0.761	0.371	0.740	0.578
11	dataLoi2008	breast cancer	+0.093	0.253	0.631	0.433	0.683	0.235	0.849	0.311	0.610
12	dataIvshina2006	breast cancer	+0.070	0.669	0.571	0.740	0.321	0.624	0.456	0.675	0.563
13	dataSabatier2010	breast cancer	+0.028	0.248	0.613	0.210	0.815	0.357	0.673	0.366	0.527
14	dataHuang2014	breast cancer	+0.004	0.644	0.540	0.575	0.420	0.789	0.221	0.816	0.505
15	dataDedeurwaerder2011	breast cancer	–0.009	0.877	0.788	0.878	0.117	0.881	0.104	0.897	0.527
16	dataWang2010	breast cancer	–0.009	0.293	0.536	0.395	0.596	0.268	0.725	0.299	0.487
17	dataLin2009	breast cancer	–0.032	0.347	0.348	0.298	0.621	0.695	0.239	0.795	0.522
18	dataKim2020	breast cancer	–0.036	0.891	0.811	0.879	0.072	0.915	0.029	0.898	0.375
19	dataMetzgerFilho2016	breast cancer	–0.055	0.832	0.723	0.839	0.094	0.835	0.109	0.851	0.464
20	dataSieber2010	colorectal cancer	+0.384	0.800	0.725	0.729	0.711	0.895	0.442	0.925	0.801
21	dataChen2020	colorectal cancer	+0.374	0.482	0.766	0.406	0.912	0.656	0.795	0.609	0.776
22	dataSmith2009b	colorectal cancer	+0.272	0.684	0.645	0.664	0.621	0.751	0.513	0.792	0.678
23	dataShinto2020	colorectal cancer	+0.240	0.356	0.696	0.540	0.739	0.339	0.887	0.492	0.723
24	dataSmith2009a	colorectal cancer	+0.225	0.666	0.620	0.672	0.549	0.673	0.557	0.708	0.650
25	dataBeauchamp2014	colorectal cancer	+0.213	0.801	0.708	0.798	0.408	0.814	0.414	0.842	0.641
26	dataMarisa2013	colorectal cancer	+0.081	0.251	0.645	0.183	0.876	0.431	0.682	0.404	0.573
27	dataGotoh2018	colorectal cancer	+0.055	0.085	0.808	0.205	0.863	0.098	0.937	0.355	0.789
28	dataStaub2009	colorectal cancer	+0.002	0.863	0.769	0.874	0.142	0.863	0.116	0.849	0.442
29	dataDelRoi2017	colorectal cancer	–0.007	0.677	0.564	0.662	0.328	0.721	0.275	0.750	0.502
30	dataSpivak2014	leukemia	+0.325	0.572	0.632	0.525	0.847	0.864	0.524	0.873	0.788
31	dataHerold2013	leukemia	+0.235	0.474	0.610	0.674	0.590	0.375	0.837	0.460	0.682
32	dataMetzeler2018	leukemia	+0.136	0.309	0.632	0.240	0.866	0.518	0.662	0.535	0.685
33	dataHerold2011	leukemia	–0.001	0.187	0.537	0.189	0.808	0.363	0.602	0.479	0.526
34	dataPotti2006	lung cancer	+0.382	0.752	0.688	0.652	0.779	0.898	0.442	0.927	0.780
35	dataKohno2011	lung cancer	+0.369	0.862	0.785	0.813	0.626	0.925	0.391	0.954	0.801
36	dataRousseaux2013	lung cancer	+0.325	0.521	0.711	0.514	0.803	0.553	0.783	0.606	0.679
37	dataSon2007	lung cancer	+0.233	0.615	0.605	0.666	0.565	0.593	0.643	0.634	0.665
38	dataBild2005	lung cancer	+0.153	0.525	0.565	0.525	0.627	0.568	0.586	0.596	0.640
39	dataBotling2012	lung cancer	+0.129	0.418	0.522	0.674	0.469	0.310	0.806	0.415	0.613
40	dataHeiskanen2015	lung cancer	+0.127	0.622	0.535	0.827	0.281	0.508	0.647	0.584	0.615
41	dataPintilie2013	lung cancer	+0.070	0.669	0.571	0.740	0.321	0.624	0.456	0.675	0.563
42	dataZChen2020	lung cancer	+0.061	0.586	0.538	0.627	0.431	0.571	0.494	0.585	0.542
43	dataTsao2010	lung cancer	+0.055	0.615	0.519	0.750	0.289	0.541	0.539	0.572	0.531
44	dataPhilipsen2010	lung cancer	+0.045	0.375	0.547	0.386	0.652	0.434	0.621	0.491	0.545
45	dataXie2011	lung cancer	–0.016	0.548	0.496	0.508	0.480	0.646	0.334	0.689	0.491
46	dataMicke2011\(^a\)	lung cancer	–0.059	0.436	0.460	0.488	0.459	0.433	0.500	0.487	0.464
47	dataVanLoo2009	lymphoma	+0.370	0.679	0.673	0.691	0.754	0.825	0.584	0.949	0.890
48	dataLenz2008	lymphoma	+0.327	0.755	0.685	0.805	0.506	0.714	0.630	0.766	0.723
49	dataIqbal2015	lymphoma	+0.168	0.525	0.565	0.646	0.520	0.473	0.698	0.537	0.623
50	dataKawaguchi2012	lymphoma	+0.015	0.215	0.606	0.251	0.762	0.291	0.706	0.458	0.550
51	dataLeich2009\(^a\)	lymphoma	–0.014	0.458	0.481	0.466	0.520	0.486	0.500	0.503	0.494
52	dataShi2010	multiple myeloma	+0.148	0.680	0.601	0.640	0.514	0.745	0.399	0.755	0.609
53	dataMulligan2007	multiple myeloma	+0.033	0.189	0.423	0.120	0.901	0.671	0.393	0.650	0.529
54	dataHiyama2009	neuroblastoma	+0.213	0.869	0.785	0.964	0.220	0.803	0.624	0.928	0.776
55	dataUehara2015	ovarian cancer	–0.070	0.800	0.678	0.752	0.116	0.887	0.032	0.889	0.412
56	dataBogunovic2009	skin cancer	+0.328	0.673	0.637	0.766	0.543	0.665	0.694	0.735	0.712
57	dataPasini2021	stomach cancer	+0.385	0.469	0.750	0.500	0.959	0.843	0.772	0.986	0.974
	average	breast cancer	+0.083	0.531	0.628	0.593	0.489	0.570	0.513	0.604	0.568
	average	colorectal cancer	+0.184	0.567	0.695	0.573	0.615	0.624	0.562	0.673	0.658
	average	leukemia	+0.176	0.422	0.621	0.440	0.745	0.549	0.637	0.604	0.668
	average	lung cancer	+0.144	0.580	0.580	0.628	0.522	0.585	0.557	0.632	0.610
	average	lymphoma	+0.173	0.526	0.602	0.572	0.612	0.558	0.624	0.643	0.656
	average	multiple myeloma	+0.091	0.435	0.512	0.380	0.708	0.708	0.396	0.703	0.569
	% sufficient scores	all datasets	33.33%	45.45%	61.82%	58.18%	21.82%	25.45%	50.91%	58.18%	58.18%
	average	all datasets	+0.138	0.545	0.620	0.595	0.546	0.593	0.556	0.646	0.619
	median	all datasets	+0.129	0.548	0.620	0.652	0.549	0.624	0.586	0.634	0.615
	min	all datasets	–0.070	0.085	0.348	0.120	0.072	0.098	0.029	0.267	0.375
	max	all datasets	+0.385	0.898	0.822	0.964	0.959	0.925	0.937	0.986	0.974

Results obtained by the Random Forests machine learning method applied to each of the 57 prognostic cancer datasets of gene expression to predict the survival or death of the patients, sorted by cancer type and Matthews correlation coefficient. We highlighted in bold all the sufficient scores: \(MCC \ge +0.2\), and F\(_1\) score, accuracy, TPR, TNR, PPV, NPV, PR AUC, ROC AUC \(\ge 0.6\). We highlighted with \(^a\) the only two datasets for which all the binary classification metrics are insufficient: dataLeich2009 and dataMicke2011. MCC, F\(_1\) score, accuracy, TPR, TNR, PPV, and NPV confusion matrix threshold cut-off: 0.5. MCC Matthews correlation coefficient. TPR true positive rate, sensitivity. TNR true negative rate, specificity. PPV positive predictive value, precision. NPV negative predictive value. PR precision recall curve. ROC receiver operating characteristic curve. AUC area under the curve. MCC has worst value –1 and best value +1. F\(_1\) score, accuracy, TPR, TNR, PPV, NPV, PR AUC, and ROC AUC have worst value 0 and best value 1. The formulas of MCC, F\(_1\) score, accuracy, TPR, TNR, PPV, NPV, PR AUC and ROC AUC can be found in the Supplementary information. % sufficient scores: percentage of datasets where the signature achieved a sufficient score (for example, our signature obtained a sufficient accuracy score on 61.82% datasets). We report additional information about these datasets in Table 1

Back to article page