Table 1 A simple model, produced by tree boosting (XGBoost), to classify human proteins as ageing-related or non-ageing-related.

From: Prediction and characterization of human ageing-related proteins by using machine learning

feature ID	description of the feature	category	score	relative frequency in ageing/non-ageing
ageing_n_0	number of ageing-related neighbours = 0	Net	−2.896	38.8/92.1
ageing_n_1	number of ageing-related neighbours = 1	Net	−2.275	15.8/5.6
ageing_n_2	number of ageing-related neighbours = 2	Net	−1.168	15.1/1.4
ageing_n_3_4	number of ageing-related neighbours = 3,4	Net	−0.744	12.8/0.6
GO:0043567	regulation of insulin-like growth factor receptor signaling pathway	BP	1.327	2.6/0.1
GO:0006979	response to oxidative stress	BP	0.9	21.7/1.4
GO:0003684	damaged DNA binding	MF	0.837	8.6/0.2
GO:0009987	cellular process	BP	0.805	99.3/70.0
GO:0005576	extracellular region	CC	0.636	21.7/8.8
GO:0065008	regulation of biological quality	BP	0.563	60.2/14.9
GO:0051276	chromosome organization	BP	0.515	14.5/1.6
GO:0032502	developmental process	BP	0.497	69.4/22.5
GO:0043066	negative regulation of apoptotic process	BP	0.474	32.9/3.5
GO:0009628	response to abiotic stimulus	BP	0.441	38.2/4.4
GO:0007169	transmembrane receptor protein tyrosine kinase signaling pathway	BP	0.413	19.1/2.1
GO:0010332	response to gamma radiation	BP	0.411	8.6/0.1
GO:0019838	growth factor binding	MF	0.405	5.3/0.4
GO:0040008	regulation of growth	BP	0.398	22.0/2.8
GO:0044710	single-organism metabolic process	BP	0.388	42.1/15.4
GO:0031325	positive regulation of cellular metabolic proc	BP	0.331	64.8/12.8
GO:0050896	response to stimulus	BP	0.288	77.3/22.8
GO:0031667	response to nutrient levels	BP	0.285	16.8/1.5
GO:0005515	protein binding	MF	0.271	75.7/24.4
GO:2000377	regulation of reactive oxygen species metabolic process	BP	0.259	13.8/0.6
GO:0051716	cellular response to stimulus	BP	0.257	62.2/11.1
GO:0005654	nucleoplasm	CC	0.235	49.7/14.1
GO:0080135	regulation of cellular response to stress	BP	0.225	27.3/2.6
GO:0048511	rhythmic process	BP	0.224	15.1/1.2
GO:0044427	chromosomal part	CC	0.197	24.0/3.4
ageing_n_5+	number of ageing-related neighbours ≥ 5	Net	0.192	17.4/0.2
GO:0003682	chromatin binding	MF	0.171	17.1/2.1
GO:0006974	cellular response to DNA damage stimulus	BP	0.167	27.6/3.1
GO:0097159	organic cyclic compound binding	MF	0.166	62.8/28.8
GO:0005739	mitochondrion	CC	0.16	20.4/6.1
GO:0019899	enzyme binding	MF	0.128	39.8/6.8
GO:0009894	regulation of catabolic process	BP	0.125	25.7/3.4

Features are listed by ID and description. Feature category can take values “Net” (Network), “MF” (Molecular Function), “CC” (Cellular Component), or “BP” (Biological Process). The table consists of only binary (true or false) features. For each protein we can compute the predicted relevance of ageing as follows: for each row of the table, we check whether the given feature is true for the protein and then we add up the corresponding scores. The larger the final sum, the more important role of a protein is predicted in the human ageing process. For example, suppose that a protein has 3 ageing-related neighbours and their UniProt record contains only two GO terms, “response to oxidative stress”, and “regulation of growth”. Then the predicted ageing relevance of that protein is − 0.744 + 0.9 + 0.398 = 0.554. Predicted scores produced by the above summation method are presented in the “Table1_pred” column of Supplementary Table S1. Scores obtained by summation are not necessarily bounded by 1. The actual output of XGBoost, which we used in the rest of the paper, was normalized to take values in [0…1]. In fact, we use the average of normalized predicted values made by several models (see the Methods). The relative frequency of features in the ageing-related and the non-ageing-related sets of proteins, a value independent of our particular model, is displayed in the last column.

Back to article page

Table 1 A simple model, produced by tree boosting (XGBoost), to classify human proteins as ageing-related or non-ageing-related.

Search

Quick links