Identification of intelligence-related proteins through a robust two-layer predictor

Figures & data

Figure 1. Pipeline of developing SVM-based models for Intell_Pred software.

Table 1. Summary of 12,669 extracted protein features that were used in SVM-based model development. More details are provided in Supplementary File S2.

	Feature name	Complete feature name	Number of features	Reference
1	AAC	Amino Acid Composition	20	[62]
2	CKSAAP	Composition of k-spaced Amino Acid Pairs	2400	[63]
3	TPC	Tri-Peptide Composition	8000	[62]
4	GAAC	Grouped Amino Acid Composition	5	[64]
5	CKSAAGP	Composition of k-spaced Amino Acid Group Pairs	150	[65]
6	GTPC	Grouped Tri-Peptide Composition	125	[65]
7	Moran	Moran correlation	240	[66–68]
8	Geary	Geary correlation	240	[68]
9	NMBroto	Normalized Moreau-Broto Autocorrelation	240	[67]
10	CTD	Composition /transition/Distribution	39	[69–72]
11	CTDC	CTDT	39	[65]
12	CTDD	CTDD	195	[65]
13	CTriad	Conjoint Triad	343	[73]
14	KSCTriad	k spaced conjoint Triad	343	[65]
15	SOCNumber	Sequence order coupling Number	60	[65]
16	QSOrder	Quasi sequence order	100	[65]
17	PAAC	Pseudo Amino Acid Composition	50	[74,75]
18	APAAC	Amphiphilic Psedudo Amino Acid Composition	80	[75,76]

Table 2. Accuracy metrics of train and test datasets for intelligence model. The optimum γ parameter value of kernel function of SVM was chosen using a grid-search technique based on five-fold cross-validation.

	Dataset 1		Dataset 2		Dataset 3		Dataset 4		Dataset 5		Average
γ	0.001		0.003		0.001		0.003		0.001		-
	Train	Test	Train	Test	Train	Test	Train	Test	Train	test	Train	Test
Accuracy (%)	89.58	88.75	88.59	87.55	89.75	87.6	88.99	87.97	88.57	86.72	89.09	87.71
Sensitivity (%)	88.06	87.80	87.20	84.73	89.50	87.10	87.72	87.10	87.38	88.70	87.97	87.08
Specificity (%)	91.43	89.74	90.25	90.91	90.22	88.14	90.45	88.89	89.92	84.92	90.45	88.52
MCC (%)	79.33	77.52	77.30	75.35	79.60	75.21	78.06	75.95	77.23	73.54	78.3	75.51

Table 3. Accuracy metrics of train and test datasets for memory model. The optimum γ parameter value of kernel function of SVM was chosen using a grid-search technique based on five-fold cross-validation.

	Dataset 1		Dataset 2		Dataset 3		Dataset 4		Dataset 5		Average
γ	0.001		0.003		0.001		0.003		0.001		-
	Train	Test	Train	Test	Train	Test	Train	Test	Train	test	Train	Test
Accuracy (%)	86.77	88.24	87.15	91.6	87.92	92.44	85.27	88.14	88.60	93.39	87.14	90.76
Sensitivity (%)	86.10	88.33	85.70	93.10	86.32	94.74	83.43	86.89	87.60	93.55	85.83	91.32
Specificity (%)	88.52	88.14	89.38	90.16	89.85	90.32	88.51	89.47	89.94	93.22	89.24	90.26
MCC (%)	74.08	76.46	74.69	83.24	75.87	84.98	71.25	76.31	77.33	86.76	74.64	81.55

Table 4. Accuracy metrics of train and test datasets for learning model. The optimum γ parameter value of kernel function of SVM was chosen using a grid-search technique based on five-fold cross-validation.

	Dataset 1		Dataset 2		Dataset 3		Dataset 4		Dataset 5		Average
γ	0.003		0.003		0.003		0.003		0.003		-
	Train	Test	Train	Test	Train	Test	Train	Test	Train	test	Train	Test
Accuracy (%)	85.30	88.43	86.76	87.6	86.96	88.33	84.27	89.17	87.78	87.5	86.21	88.2
Sensitivity (%)	83.15	86.15	84.19	83.82	84.28	84.85	81.09	85.07	85.74	85.71	83.69	85.12
Specificity (%)	88.63	91.07	90.05	92.45	90.67	92.59	89.37	94.34	90.36	89.47	89.81	91.98
MCC (%)	71.19	77.01	73.89	75.69	74.43	77.05	69.50	78.87	75.83	75.09	72.96	76.74

Table 5. Summary results of evaluation of the 10,000 new negative protein sequences by Intell_Pred.

	Score* >0.5		Score >0.75	Score >0.90
	Number	Percent	Number	Percent	Number	Percent
Intelligence	1058	10.58	612	6.12	347	3.47
Learning	844	8.44	587	5.87	308	3.08
Memory	815	8.15	484	4.84	219	2.19

* Score indicates the estimated score for a protein to be intelligence-related in the first layer and its score in the second layer to be involved in learning or memory. For example, out of 10,000 negative proteins, only 1058 proteins were detected to be intelligence-related proteins. Higher scores are more reliable.

Table 6. Summary results of evaluation of the new annotated protein sequences related to intelligence in Uniprot database by Intell_Pred (169 protein sequences were evaluated including 85 and 84 proteins belong to memory and learning classes, respectively) .

	Score* >0.5		Score >0.75		Score >0.90
	Number	Percent	Number	Percent	Number	Percent
Intelligence	163	96.45	159	94.08	137	81.07
Learning	81**	96.42	77	91.66	71	84.52
Memory	81 ***	95,329	79	92.94	69	81.17

* Score indicates the estimated score for a protein to be intelligence-related in the first layer and its score in the second layer to be involved in learning or memory. For example, out of 169 evaluated proteins, 163 proteins were detected to be intelligence-related proteins. Higher scores are more reliable. ** Out of 84 proteins that were belonged to learning class. *** Out of 85 proteins that were belonged to memory class.

Table 7. Summary results of evaluation of the plant candidate proteins using Intell_Pred.

Intelligent attribute	Related proteins	Number of proteins	Intelligence			Learning			Memory
			≥0.5*	≥0.75	≥0.9	≥0.5	≥0.75	≥0.9	≥0.5	≥0.75	≥0.9
Associate learning in pea	Phototropins	18	17	16	13	17	15	9	17	12	4
Semimonastic movements	Pulvinus activation, … **	52	44	33	21	42	35	20	38	25	10

*Predicted scores. ** Turgor pressure, Leaf movement, Cytochalasin b-related protein, Motor cells related proteins, Charge amplification related protein, Osmotic pressure.

Bhasin M, Raghava GP. Classification of nuclear receptors based on amino acid composition and dipeptide composition. J Biol Chem. 2004;279:23262–23266.

 PubMed Web of Science ®Google Scholar

Chen K, Jiang Y, Du L, et al. Prediction of integral membrane protein type by collocated hydrophobic amino acid pairs. J Comput Chem. 2009;30:163–172.

 PubMed Web of Science ®Google Scholar

Lee T-Y, Lin Z-Q, Hsieh S-J, et al. Exploiting maximal dependence decomposition to identify conserved motifs from a group of aligned signal sequences. Bioinformatics. 2011;27:1780–1787.

 PubMed Web of Science ®Google Scholar

Chen Z, Zhao P, Li F, et al. iFeature: a python package and web server for features extraction and selection from protein and peptide sequences. Bioinformatics. 2018;34:2499–2502.

 PubMed Web of Science ®Google Scholar

Sokal RR, Thomson BA. Population structure inferred by local spatial autocorrelation: an example from an Amerindian tribal population. Am J Phys Anthropol. 2006;129:121–131.

 PubMed Web of Science ®Google Scholar

Horne DS. Prediction of protein helix content from an autocorrelation analysis of sequence hydrophobicities. Biopolymers. 1988;27:451–477.

 PubMed Web of Science ®Google Scholar

Shen J, Zhang J, Luo X, et al., 2007. Predicting protein–protein interactions based only on sequences information. Proceedings of the National Academy of Sciences 104, 4337–4341.

 Google Scholar

Chou K-C. Prediction of protein cellular attributes using pseudo-amino acid composition. Proteins: structure. Funct Bioinf. 2001a;43:246–255.

 PubMed Web of Science ®Google Scholar

Chou K-C. Using amphiphilic pseudo amino acid composition to predict enzyme subfamily classes. Bioinformatics. 2004;21:10–19.

 PubMed Web of Science ®Google Scholar

Chou K-C. Using subsite coupling to predict signal peptides. Protein Eng Des Sel. 2001b;14:75–79.

 Google Scholar