Fig 1.
The workflow pipeline followed to build the disease similarity network (DSN) of high altitude related diseases.
Step 1: PubMed Central was queried for high altitude related abstracts. Step 2: All high altitude abstracts were collected and redundancies were removed. Step 3: The Biomedical concept annotation system (BeCAS) was used to extract the high altitude related diseases/disorders and chemicals from the collected abstracts. Step 4: The relations among diseases and chemicals from Comparative Toxicogenomics Database (CTD) were used to construct the Disorder-Chemical Network. Step 5: Disease ontology based semantic similarity-based score was used to construct the Disease similarity network (DSN).
Fig 2.
The pipeline for prediction of comorbid seasonal diseases from DSN.
The pipeline starts from the DSN corresponds to step 5 in Fig 1. Step1: the DSN was clustered using four community clustering algorithms of igraph program. Step 2: These community clusters were overlapped to predict the 12 diseases (core-DSN). Step 3: The monthly Google RSV were collected for the 12 diseases with obesity as the benchmark disease. Step 4: The low RSV diseases were removed, and the remaining diseases were subjected to seasonal Mann-Kendall and Dynamic time wrapping methods. Step 5: Based on their seasonal trend p-values and shape distance matrix clustering the diseases were divided into two groups severe (high) and moderate (low). Step 6: Their seasonal trends were further verified using seasonal decomposition analysis by TABTS and LOESS. The periodicity was analyzed using Autocorrelation and Fourier series analysis.
Fig 3.
Methods for disease-drug network-based selection for the most relevant high-altitude diseases.
(a) Link prediction model predict the relationship between high altitude related drugs (yellow) and diseases (orange) based on the clinically reported drug-disease pairs association were obtained from Comparative Toxicogenomics Database (CTD). The node size of each disease term in the network corresponds to their frequency of occurrence in the high altitude related abstracts. (b) The outlier plot showed diseases have the high term frequency (n>10) called as bottleneck diseases.
Fig 4.
Disease ontology (DO) based semantic similarity disease network (DSN) of high altitude.
The disease (red colour square shape) pairs showing the >90 percentile (outliers) literature frequency were used in the construction. The average degree (number of links with other diseases) of all diseases in the disease network is 0.1418 (marked as gray lines). The edge thickness represents the SS score between two diseases. Note that the bottleneck diseases of high altitude in the network are in square shapes rather than circle otherwise.
Fig 5.
The DSN subjected to the four community detection algorithms based on the (a) edge betweenness (EB), (b) fast greedy (FG), (c) spin glass (SG) and (d) walk trap (WT) available in the “igraph” package. Here, we clearly see that among the six bottleneck diseases (square shape edges) only hypertension community (green colour) and obesity community (magenta colour) are tightly maintained by the four community detection algorithms (encircled). Please note that in all the community detection algorithms, the hypertension community associated diseases (acidosis, hypocholesterolemia, hyperhomocysteinemia, and hyperglycemia) formed a major community cluster. Similarly, the obesity community diseases (esophagitis and fibrosis) formed a short community cluster. Whereas, other bottleneck diseases not able to maintain a separate community cluster. Moreover, the overall inter community cluster interactions (red colour edges) are more than intra community cluster interactions (gray colour edges). Visualization of the network was done using cytoscape (Shannon et al., 2003).
Fig 6.
The left-hand side is the DSN of high altitude marked with six bottleneck diseases (square, red colour) and six community diseases (square, green colour).
The overall semantic similarity (SS) average score of the DSN is 0.14188. The right-hand side is core DSN network of 12 diseases (six bottleneck and six community diseases). The overall semantic similarity score of the core DSN is 0.19624 and named as “highly comorbid diseases of high altitude”.
Fig 7.
The monthly widespread google trend relative search volume (RSV) collected for the “highly comorbid diseases of high altitude (core DSN)” with obesity (dark blue colour) as the benchmark disease.
The average RSV of the core-DSN was diseases more than 20 for the entire period were boxed. These seven diseases were named as “severe seasonal comorbid lifestyle diseases (SCLD)”.
Fig 8.
Widespread seasonal sensitive comorbid diseases.
(a) Seasonal Mann-Kendall p-values of seasonal sensitive life style diseases represented as heat map for the entire period (Jan 2004 to Dec 2016). The hierarchical clustering clearly divided them into diseases with (red box) and without (blue box) season trend. (b) Shape based distance matrix scores among the major highly comorbid diseases of high altitude represented as heat map for the entire period (Jan 2004 to Dec 2016). The hierarchical clustering clearly divided them into diseases with (red box) and without (blue box) seasonal pattern matching. Please note that the four diseases (obesity, asthma, hypertension, and fibrosis) in the red boxes have similar seasonal trends and seasonal patterns named as “Group1” diseases (S1 Table).
Fig 9.
Seasonal and trend decomposition using TBATS for the four group 1 diseases for the monthly collected RSV from Jan 2004–Dec 2016 with obesity as the benchmark disease.
Raw observed data were displayed in the top panel as averaged values for all transect points with monthly sampling frequency, followed by level, trend, and seasonal components 1,2, and 3. Scale differs for each of the components, and so relative magnitude was indicated by the gray bars on the left side of the panels.
Fig 10.
Seasonal and trend decomposition using loess (STL) for the four group 1 diseases for the monthly collected RSV from Jan 2004 –Dec 2016 with obesity as the benchmark disease.
Raw data are displayed in the top panel as averaged values for all transect points with monthly sampling frequency, followed by seasonal, trend, and residual components. Scale differs for each of the components, and so relative magnitude was indicated by the gray bars on the right side of the panels.
Fig 11.
Autocorrelation and periodicity of four group 1 diseases for the monthly collected RSV from Jan 2004–Dec 2016 with obesity as the benchmark disease.
Observed data is showing a cyclic pattern in autocorrelation above significant line (dotted) and six months periodicity in all and three months in asthma as well as hypertension (periodogram).
Table 1.
Widespread monthly RSV of GT time series seasonal analysis of the core DSN diseases (Jan 2004 to Dec 2016) with obesity as the reference term.
(Complete raw data is given in S1 Table).
Table 2.
Widespread monthly RSV of GT time series seasonal analysis of the core DSN diseases (Jan 2004 to Dec 2016) without the reference term.
(Complete raw data is given in S2 Table).
Fig 12.
Season wise BMA of group 1 diseases for USA (Red in color) and NZL (Blue in color).
The X-axis time period (2004–2016) of each year divided into the window size of 6 months and labelled as April (APR) to September (SEP) and October (OCT) to March (MAR). The Y-axis marked with BMA of Group 1 diseases for the 6 months window size. Please note that USA and NZL seasonal search patterns were juxtaposed (vertical gray bars).
Table 3.
The monthly RSV of GT time series seasonal analysis of USA (NH) and NZL (SH) for Group 1 diseases (Jan 2004 to Dec 2016) with obesity as the reference term.
(Complete raw data is given in S3 Table).
Table 4.
Widespread weekly time series seasonal analysis of life style diseases (Jan 2004 to Dec 2016).
(Complete raw data is given in S4 Table).
Fig 13.
Network view of comorbid association reported in the literature among different diseases of the core DSN.
The connection between the different diseases nodes were labelled with their corresponding PubMed Identification number (PMIDs). The complete reference of the labelled PMIDs is given in the S5 Table.
Table 5.
List of clinical studies (PMIDs) reported the seasonal and life style factors impact on etiology of the core DSN.
Please note that this list is not exhaustive and only includes some of the frequently cited reference PMIDs. The citation numbers of each PMIDs as on June 2018 from PubMed is given in the bracket.