HLPI-Ensemble: Prediction of human lncRNA-protein interactions based on ensemble strategy

Figures & data

Table 1. The prediction results of SVM, RF and XGB by Pse-in-One feature combinations in NPTest1.

Algorithm	LncRNA feature	Protein feature	AUC	ACC	REC	SPEC	PPV	F1 score
SVM	Kmer	Kmer	0.9540	0.9229	0.9750	0.8703	0.8837	0.9271
SVM	Kmer	AC	0.9523	0.9227	0.9750	0.8699	0.8834	0.9269
SVM	Kmer	PC-PseAAC-General	0.9574	0.9268	0.9669	0.8862	0.8958	0.9300
SVM	DAC	Kmer	0.9572	0.9229	0.9775	0.8677	0.8820	0.9273
SVM	DAC	AC	0.9565	0.9209	0.9698	0.8714	0.8841	0.9250
SVM	DAC	PC-PseAAC-General	0.9595	0.9188	0.9504	0.8870	0.8948	0.9217
SVM	PC-PseDNC-General	Kmer	0.9565	0.9222	0.9735	0.8703	0.8836	0.9264
SVM	PC-PseDNC-General	AC	0.9541	0.9222	0.9739	0.8699	0.8833	0.9264
SVM	PC-PseDNC-General	PC-PseAAC-General	0.9573	0.9249	0.9617	0.8877	0.8965	0.9280
RF	Kmer	Kmer	0.9630	0.9222	0.9750	0.8688	0.8826	0.9265
RF	Kmer	AC	0.9613	0.9214	0.9750	0.8673	0.8814	0.9258
RF	Kmer	PC-PseAAC-General	0.9678	0.9264	0.9709	0.8814	0.8923	0.9299
RF	DAC	Kmer	0.9649	0.9261	0.9753	0.8762	0.8885	0.9299
RF	DAC	AC	0.9629	0.9242	0.9775	0.8703	0.8840	0.9284
RF	DAC	PC-PseAAC-General	0.9671	0.9266	0.9717	0.8810	0.8920	0.9301
RF	PC-PseDNC-General	Kmer	0.9634	0.9224	0.9739	0.8703	0.8836	0.9265
RF	PC-PseDNC-General	AC	0.9612	0.9218	0.9757	0.8673	0.8815	0.9262
RF	PC-PseDNC-General	PC-PseAAC-General	0.9682	0.9273	0.9720	0.8821	0.8930	0.9308
XGB	Kmer	Kmer	0.9670	0.9285	0.9654	0.8911	0.8996	0.9314
XGB	Kmer	AC	0.9657	0.9262	0.9665	0.8855	0.8952	0.9295
XGB	Kmer	PC-PseAAC-General	0.9654	0.9275	0.9654	0.8892	0.8981	0.9305
XGB	DAC	Kmer	0.9669	0.9285	0.9735	0.8829	0.8937	0.9319
XGB	DAC	AC	0.9664	0.9266	0.9709	0.8818	0.8926	0.9301
XGB	DAC	PC-PseAAC-General	0.9685	0.9294	0.9742	0.8840	0.8947	0.9328
XGB	PC-PseDNC-General	Kmer	0.9674	0.9285	0.9654	0.8911	0.8996	0.9314
XGB	PC-PseDNC-General	AC	0.9660	0.9268	0.9617	0.8914	0.8996	0.9296
XGB	PC-PseDNC-General	PC-PseAAC-General	0.9688	0.9288	0.9647	0.8926	0.9008	0.9317

Table 2. The performance of average strategy and linear strategy that verified by 100 times 10-fold CV in NPTest2.

Model	Ensemble strategy	AUC	AUC sd	ACC	REC	SPEC	PPV	F1 score
HLPI-SVM Ensemble	average	0.9557	0.0087	0.9094	0.9343	0.8852	0.8880	0.9105
HLPI-SVM Ensemble	linear	0.9535	0.0092	0.9177	0.9664	0.8703	0.8789	0.9205
HLPI-RF Ensemble	average	0.9593	0.0082	0.8993	0.8916	0.9067	0.9030	0.8971
HLPI-RF Ensemble	linear	0.9629	0.0078	0.9229	0.9590	0.8878	0.9229	0.9246
HLPI-XGB Ensemble	average	0.9644	0.0073	0.9100	0.9077	0.9122	0.9096	0.9085
HLPI-XGB Ensemble	linear	0.9631	0.0075	0.9171	0.9598	0.87563	0.8825	0.9194

Table 3. The performance of all sub-models in external validation.

Model	LncRNA feature	Protein feature	AUC	ACC	REC	SPEC	PPV	F1 score	TP	FN	FP	TN
SVM	Kmer	Kmer	0.7351	0.5687	0.9629	0.1744	0.5384	0.6906	2628	101	2253	476
SVM	Kmer	AC	0.7172	0.5381	0.9728	0.1033	0.5203	0.6780	2655	74	2447	282
SVM	Kmer	PC-PseAAC-General	0.7326	0.6357	0.9278	0.3437	0.5857	0.7180	2532	197	1791	938
SVM	DAC	Kmer	0.7243	0.5122	0.9849	0.0395	0.5063	0.6688	2688	41	2621	108
SVM	DAC	AC	0.7495	0.5461	0.9589	0.1333	0.5252	0.6787	2617	112	2365	364
SVM	DAC	PC-PseAAC-General	0.7088	0.6416	0.8603	0.4228	0.5985	0.7059	2348	381	1575	1154
SVM	PC-PseDNC-General	Kmer	0.7446	0.5390	0.9703	0.1077	0.5209	0.6779	2648	81	2435	294
SVM	PC-PseDNC-General	AC	0.7340	0.5388	0.9681	0.1095	0.5208	0.6773	2642	87	2430	299
SVM	PC-PseDNC-General	PC-PseAAC-General	0.7433	0.6299	0.9234	0.3363	0.5818	0.7138	2520	209	1811	918
RF	Kmer	Kmer	0.8066	0.5344	0.9791	0.0897	0.5182	0.6777	2672	57	2484	245
RF	Kmer	AC	0.8089	0.5153	0.9893	0.0414	0.5079	0.6712	2700	29	2616	113
RF	Kmer	PC-PseAAC-General	0.8026	0.6227	0.9377	0.3078	0.5753	0.7131	2559	170	1889	840
RF	DAC	Kmer	0.7974	0.5732	0.9582	0.1883	0.5414	0.6918	2615	114	2215	514
RF	DAC	AC	0.7935	0.5472	0.9739	0.1205	0.5255	0.6826	2658	71	2400	329
RF	DAC	PC-PseAAC-General	0.7956	0.6143	0.9428	0.2858	0.5689	0.7096	2573	156	1949	780
RF	PC-PseDNC-General	Kmer	0.8106	0.5489	0.9721	0.1256	0.5264	0.6830	2653	76	2386	343
RF	PC-PseDNC-General	AC	0.8071	0.5164	0.9890	0.0439	0.5084	0.6716	2699	30	2609	120
RF	PC-PseDNC-General	PC-PseAAC-General	0.8036	0.6271	0.9369	0.3173	0.5785	0.7153	2557	172	1863	866
XGB	Kmer	Kmer	0.8032	0.6423	0.9391	0.3455	0.5893	0.7242	2563	166	1786	943
XGB	Kmer	AC	0.7930	0.6452	0.9292	0.3613	0.5926	0.7237	2536	193	1743	986
XGB	Kmer	PC-PseAAC-General	0.7971	0.6582	0.9270	0.3895	0.6029	0.7306	2530	199	1666	1063
XGB	DAC	Kmer	0.7938	0.6201	0.9483	0.2920	0.5725	0.7140	2588	141	1932	797
XGB	DAC	AC	0.8205	0.6311	0.9501	0.3122	0.5800	0.7203	2593	136	1877	852
XGB	DAC	PC-PseAAC-General	0.8046	0.6192	0.9468	0.2916	0.5720	0.7132	2584	145	1933	796
XGB	PC-PseDNC-General	Kmer	0.7875	0.6476	0.9380	0.3572	0.5934	0.7269	2560	169	1754	975
XGB	PC-PseDNC-General	AC	0.7776	0.6526	0.9201	0.3851	0.5994	0.7259	2511	218	1678	1051
XGB	PC-PseDNC-General	PC-PseAAC-General	0.7914	0.6575	0.9322	0.3829	0.6017	0.7313	2544	185	1684	1045

Table 4. The performance of all HLPI-Ensemble models in external validation.

Model	AUC	ACC	REC	SPEC	PPV	F1 score	TP	FN	FP	TN
HLPI SVM-Ensemble	0.7559	0.7039	0.8805	0.5272	0.6506	0.7483	2403	326	1290	1439
HLPI RF-Ensemble	0.7998	0.6844	0.8995	0.4694	0.6290	0.7403	2455	274	1448	1281
HLPI XGB-Ensemble	0.8192	0.7643	0.8292	0.6995	0.7340	0.7787	2263	466	820	1909

Figure 1. The ROC curves of HLPI-SVM Ensemble, HLPI-RF Ensemble, HLPI-XGB Ensemble, RPISeq-SVM, RPISeq-RF and lncPro are expressed in red, brown, green, blue, navy blue and pink, respectively. The maximum area under the curve (AUC) represents the best performance of the model.

Table 5. The performance of HLPI-Ensemble, RPI-Pred, lncPro, RPISeq-SVM and RPISeq-RF in lncRNome database.

Model	AUC	ACC	REC	SPEC	PPV	F1 score	TP	FN	FP	TN
HLPI SVM-Ensemble	0.7559	0.7039	0.8805	0.5272	0.6506	0.7483	2403	326	1290	1439
HLPI RF-Ensemble	0.7998	0.6844	0.8995	0.4694	0.6290	0.7403	2455	274	1448	1281
HLPI XGB-Ensemble	0.8192	0.7643	0.8292	0.6995	0.7340	0.7787	2263	466	820	1909
RPI-Pred	—	0.4541	0.8732	0.0351	0.4750	0.6153	2383	346	2633	96
RPISeq-SVM	0.5081	0.0218	0.8534	0.1623	0.5046	0.6342	2329	400	2286	443
RPISeq-RF	0.5217	0.5027	0.9849	0.0205	0.5013	0.6645	2688	41	2673	56
lncPro	0.5600	0.5271	0.8490	0.2052	0.5164	0.6422	2317	412	2169	560

Figure 2. The process of constructing benchmark dataset.

Figure 3. The construction process of HLPI-Ensemble model.

Table 6. The coefficients of HLPI-Ensemble models trained by linear ensemble strategy.

HLPI-Ensemble model	${\beta }_0$	${\beta }_1$	${\beta }_2$	${\beta }_3$	${\beta }_4$	${\beta }_5$	${\beta }_6$	${\beta }_7$	${\beta }_8$	${\beta }_9$
HLPI-SVM Ensemble	4.206	−0.497	−1.095	−1.844	−8.087	6.045	−2.911	14.73	−12.53	−1.164
HLPI-RF Ensemble	4.352	6.161	−3.442	−4.094	−2.732	2.522	−4.186	7.761	−3.863	−6.288
HLPI-XGB Ensemble	4.328	0.5581	0.3794	−1.700	0.0096	−1.247	−3.704	1.295	−2.886	−0.1375