Page 18 - GPD-2-3

Gene & Protein in Disease Signatures construction strategies for TC

TC patients are surgical resection, radioiodine, and systemic total of 42 studies were collected (Figure 1), 45 prognostic
therapy according to different pathological types [9,15,16] . signatures for TC were summarized, and data types include
Therefore, if risk factors of TC progression could be precisely mRNA, miRNA, and lncRNA. These prognostic signatures
predicted, early intervention and targeted therapy might were associated with overall survival (OS), disease-
prevent the deterioration of TC, which would improve free survival (DFS), recurrence-free survival (RFS), or
the prognosis of TC patients [17,18] . Recently, researchers progression-free interval (PFI) of TC patients. After
have identified several signatures of prognostic values for analyzing and summarizing these signatures, we proceeded
TC patients and developed risk prediction models of TC with the following works: (1) sorting out data sources and
progression to predict survival outcomes, classify patients by applications; (2) categorizing these prognostic signatures;
risk stratification, and guide treatments for TC [19-21] . (3) summarizing three main strategies for constructing
This retrospective review discussed the previously signatures; (4) summing up the verification methods
reported signatures of TC, concluded three main strategies of signatures; (5) summarizing nomograms of these
for signature construction related to TC prognosis, and prognostic signatures; and (6) screening overrepresented
summarized the methods of verifying signatures after prognostic genes.
integrated analysis. 3.1. Summary of the sources of data for signature

2. Methods construction
In general, for the establishment of prognostic signatures,
2.1. Data collection and selection
the most pervasive source of data acquisition to download
To estimate previously reported prognostic signatures for expression profiling and patients’ clinical information was
TC, a total of 150 studies were obtained by searching the from the TCGA or GEO database. Besides, some studies
PubMed database with the keywords “prognostic signature obtained data from some specific databases, for instance,
AND thyroid cancer.” Prognostic signatures derived from the Human Autophagy Database (HADb) [22,23] , the
these studies included four categories: mRNA signatures, Immunology Database, and the Analysis Portal (ImmPort)
non-coding RNAs (ncRNAs) signatures, genomic signatures, database [24-26] . These databases were typically rich in
and signatures related to biological functions. Selection genes and molecules associated with specific biological
of studies was then conducted according to the consistent functions. In addition, some researchers obtained data
exclusion criteria listed below: (1) review; (2) not focused on from array or sequencing results of a certain number
TC; (3) not in English; (4) only one molecule mentioned; and of qualified thyroid cancer and normal tissue samples
(5) no prognostic signature constructed/no survival analysis derived from TC patients [21,27-33] . For instance, in addition
mentioned. As a result, a total of 42 studies correlated with to obtaining 165 transcriptome data and 125 PTC patients’
TC prognosis were included in the study (Figure 1). clinical data from the Nucleotide Archive database, Teng
et al. also included 11 patients who had undergone total
2.2. Evaluation of signatures through meta-analysis
thyroidectomy in Beijing Cancer Hospital . The data
[21]
To evaluate the prognostic abilities of different signatures obtained from different approaches or platforms were not
for TC, survival data from the training set were collected only the first step to constructing molecular signatures
and summarized, including (1) survival curves and risk but also laid the foundation for researchers to carry out
scores, (2) receiver operating characteristic (ROC) plots, subsequent works.
(3) univariate and multivariate Cox regression analysis, and
(4) nomogram analysis. To visually estimate the abilities 3.2. Classification of prognostic signatures for TC
of risk stratification of signatures for TC patients, values Derived from different data types, these signatures were
of hazard ratios (HR) and 95% confidence intervals (CI) divided into four categories. Signatures of 19 studies
were extracted from multivariate regression analysis data were identified by analyzing mRNA data [19,20,34-40] and
of these studies and described as a forest plot (Figure 2) ncRNA data involving long ncRNAs (lncRNAs) [33,41-43] and
using GraphPad Prism 8.0.2. As described, 28 prognostic microRNAs (miRNAs) [29,44] . Signatures of 11 studies were
signatures estimated by meta-analysis presented great associated with specific biological functions, including
prediction performance that most high-risk TC patients immune-related genes (IRGs) [24-26] , autophagy-related
had poorer survival rates than low-risk patients (Figure 2). genes (ARGs) [24-26] , epithelial-mesenchymal transition
(EMT)-related genes , RBPs-associated genes , and
[45]
[30]
3. Results ferroptosis-related genes [27,31,46] . In addition, signatures
Comprehensive signature information for TC prognosis of eight studies were related to mutation or methylation,
was described in Tables 1-4 and Figure 2. By screening, a including methylation-driven genes and regulators [48,49] .
[47]

Volume 2 Issue 3 (2023) 2 https://doi.org/10.36922/gpd.1138