We modeled the epigenetic interactions between TMAO and diseases on both GDN_OMIM and GDN_GWAS by inserting TMAO into these two disease networks. We obtained human genes associated with TMAO from STITCH, a publicly available database of known and predicted interactions of chemicals and proteins [19]. Currently, STITCH contains interactions for between 300,000 small molecules and 2.6 million proteins from 1,133 organisms, with each interaction associated with a score measuring the evidence of the association. In STITCH, TMAO is associated with a total of 553 genes from 932 species, including 54 genes from humans. Table 1 shows ten human genes associated with TMAO.

We first inserted a pseudo-node representing TMAO into GDNs. This node was then connected to disease nodes on GDNs if TMAO-associated genes interact with disease-associated genes. The edge weights were determined by the numbers of interacting genes between the newly inserted node (T) and existing disease nodes (D_j) and is defined as: WDj= ∑k=1nGTk ∑l=1mGjl where G_T k is a gene associated with TMAO, G_j l is a gene associated with D_j , and G_T k is the same as or interacts with G_j l. The intuition is that: if TMAO-associated genes participate in the same pathways as disease-associated genes, we can hypothesize that TMAO may be involved in disease pathogenesis. The degree of relatedness between TMAO and diseases was determined by the numbers of interacting gene pairs. After inserting the node "TMAO", we re-normalized the matrices represented by EINs and applied an existing network-based ranking algorithm to find diseases that share high genetic similarities with TMAO (Figure 1). The network-based ranking algorithm was used to find diseases that are related to TMAO both directly and indirectly by taking into account of inter-relationships among diseases. After this step, we created two EINs: the EIN_OMIM was created based on GDN_OMIM and the EIN_GWAS was created based on GDN_GWAS.

We then developed a network-based ranking algorithm to prioritize diseases on EINs based on their genetic commonalities with TMAO. We retargeted the TopicSensitive PageRank (TSPR) algorithm to rank similar diseases for a given input (TMAO in our study). TSPR is a context-sensitive ranking algorithm for web searches developed by Taher Haveliwala [20]. Versions of this approach have been used in prioritizing disease genes using networks consisting of same node types (i.e. diseases or genes) [21,22]. In this study, we applied the same algorithm to a heterogeneous network consisting of disease nodes and chemical nodes (TMAO in this study) in order to find TMAO-related diseases. The iterative networkbased ranking algorithm in finding similar diseases to a given input is defined as: p_t+1 = (1 − r)M p_t + rp₀, wherein M is the column-normalized adjacency matrix of EINs, γ is a preset probability of restarting from the initial seed node (γ = 0.1 in this study), and p^t is a vector in which the i_th element holds the normalized ranking score of disease i at t_th iteration. The initial probability vector p⁰ contains normalized probability for input. In our study, p⁰ contains TMAO, with a probability of 1.0. Diseases are then ranked according to the value in the steady-state probability vector, which is obtained by iterating the algorithm until the change between p^t+1 and p^t is less than 10⁶.

Recent studies indicate that high levels of TMAO in the blood are associated with an increased risk of cardiovascular diseases [11-14]. We examined the rankings of cardiovascular diseases and their major risk factors, including high blood cholesterol and triglyceride, high blood pressure, diabetes, and obesity, among diseases retrieved from EINs using TMAO as seed. As positive controls, these diseases are expected to rank highly among TMAO-related diseases.

In order to provide evidence supporting our hypothesis that TMAO may be involved in CRC pathogenesis, we tested whether CRC would rank highly among TMAOrelated diseases retrieved from both EINs. High rankings of CRC would imply that TMAO and CRC share high genetics and that TMAO might be associated with CRC carcinogenesis.

To better understand TMAO-related diseases, we determined the kinds of diseases that were enriched among top-ranked diseases retrieved from EINs. We classified diseases into different categories using the 10th revision of the International Statistical Classification of Diseases and Related Health Problems (ICD10), a disease classification scheme designated by the World Health Organization (WHO) [23]. The ICD10 includes 22 highest-level disease classes. We used 16 of the 22 chapters and excluded six non-specific disease classes. Since the terms used in ICD10 may be different from those used in EINs, we mapped disease terms in ICD10 to their synonyms through the unified medical language system (UMLS) unique concept identifiers [24]. Disease chapters and the numbers of diseases in each chapter are listed in Table 2.

Since EIN_OMIM contains 4,848 disease nodes and EIN_GWAS contains 882 nodes, we performed disease class enrichment analysis on TMAO-related diseases retrieved from EIN_OIMIM only. For diseases ranked at 10 different ranking cutoffs (top 10%, 20%, . . . 100%), we calculated percentages of the sixteen ICD10 disease classes among them.

In order to gain insights into common mechanistic relationships shared between TMAO and CRC, we identified and ranked genetic pathways linking them (Figure 2). Functions of highly enriched pathways might provide insights into common molecular mechanisms linking TMAO to CRC. TMAO is associated with 54 human genes based on the STITCH database. CRC is associated with 65 genes according to the GWAS catalog and 53 genes according to the OMIM database. There is no direct overlap between CRC-associated genes from OMIM and those from the GWAS catalog. In addition, there is no overlap between TMAO-associated genes and CRC-associated genes. We analyzed gene-associated pathways using the pathway data (a total of 10,295 pathways and gene sets) from the Molecular Signatures Database (MSigDB), a collection of annotated genetic pathways or gene sets from multiple sources [25]. We ranked these pathways based on the numbers of genes associated with TMAO or CRC: Rpathway= ∑i=1nGi, where G_i is a TMAOor CRC-associated gene that a given pathway contains. We then identified pathways that contain both TMAOand CRC-associated genes and ranked them base on the number of interacting gene-gene pairs between TMAO genes and CRC genes: Rcommon_pathway= ∑i=1nGi ∑j=1mGj where G_i is a TMAO-associated gene and G_j is a CRC-associated gene.

Recent studies indicate that high levels of TMAO in the blood are associated with an increased risk of CVDs. Our results demonstrated that CVDs as well as their major risk factors, including high blood cholesterol and triglyceride, high blood pressure, diabetes, and obesity, were ranked highly among TMAO-related diseases retrieved from both EIN_OMIM and EIN_GWAS. We retrieved a total of 878 diseases/traits from EIN_GWAS, among which obesity-related traits was ranked at top 1 (top 0.11%) and coronary heart disease at top 7 (top 0.80%). Other CVDrelated risk factors, including HDL cholesterol (top 1.13%), type 2 diabetes (top 1.48%), LDL cholesterol (top 1.82%), and metabolic syndrome (top 3.64%) were also ranked highly (Table 3).

We retrieved a total of 4,732 diseases from EIN_OMIM using TMAO as input. Similar to results based on EIN_OMIM, CVDs and their major risk factors, including myocardial infarction (top 0.23%), ventricular tachycardia (0.25%), diabetes mellitus, noninsulin-dependent (top 0.32%), and coronary artery disease, susceptibility to (top 0.51%), were ranked highly. Even though the diseases from EIN_GWAS (mainly common complex diseases) and from EIN_OMIM (mainly rare Mendelian disorders) are largely complementary, the high rankings of CVDs and their major risk factors among TMAO-related diseases retrieved from both networks confirmed recent studies and validated our network-based approach in finding TMAO-related diseases.

Table 4 shows top ten TMAO-related diseases/traits retrieved from EIN_GWAS and from EIN_OMIM. Colorectal cancer was ranked at top 1 (top 0.02%) among 4732 diseases retrieved from EIN_OMIM and at top 10.6% among the 882 retrieved diseases/traits. The high rankings of colorectal cancers based on both networks provided strong evidence supporting our hypothesis that TMAO may be involved in colorectal cancer pathogenesis. Since the GWAS catalog mainly contains common complex diseases/traits and the OMIM database mainly contains Mendelian diseases, the top-ranked diseases retrieved from the two disease networks are quite different. The top ten TMAO-related diseases/traits from EIN_GWAS included several CVD-related risk factors, including obesity-related traits, metabolite levels, coronary heart disease, metabolic traits, and HDL cholesterol. Interestingly, three autoimmune diseases including inflammatory bowel disease, multiple sclerosis, and Crohn's disease were also ranked within top ten. The relationship between TMAO and autoimmune diseases warrants further investigation.

Strikingly, among top ten TMAO-related diseases retrieved from EIN_OMIM, seven are cancers, including CRC, breast cancer, gastric cancer and leukemia. Because of the strong (causal) disease-gene associations in the large OMIM database, the observed strong relationship between TMAO and cancers implies that TMAO might be genetically involved in not only CRC but also cancers in general, which we further confirmed in the next section.

We examined the distributions of sixteen disease classes among 4,732 TMAO-related diseases retrieved from EIN_OMIM at 10 different ranking cutoffs (top 10%, 20%,

. . . 100%). Among the sixteen disease classes, only two disease classes were enriched among top-ranked TMAO-related diseases: Neoplasms and Endocrine, nutritional and metabolic diseases (Figure 3). For example, 11.79% of the top 10% ranked diseases were neoplasms, representing a significant 211.9% increase as compared to 3.79% of neoplasms among all retrieved diseases. Similarly, a total of 18.74% of the top 10% ranked diseases were metabolic diseases, representing a 47.8% increase as compared to 12.68% among all retrieved diseases. Given the limited number of diseases contained in the GWAS catalog, we did not perform disease enrichment analysis on diseases/traits retrieved from EIN_GWAS.

We demonstrated that CRC was highly related to TMAO in afore-mentioned sections. We next investigated common genetic pathways that are involved in both TMAO and CRC. The 54 TMAO-associated human genes are involved in a total of 170 pathways. The 53 CRC genes based on OMIM genetics are involved in 503 pathways and the 65 CRC genes based on GWAS studies are associated with 182 pathways. Although no specific genes are shared between TMAO and CRC, many common genetic pathways are associated with both: 52 common pathways between TMAO and CRC based on OMIM genes and 39 common pathways based on GWAS genetics (Table 5).

Even though there is no overlap between the 53 CRC-associated genes identified from OMIM and the 65 CRC-associated genes identified from the GWAS catalog, these genes shared 118 pathways, which we used to identify genetic pathways linking CRC and TMAO. We found that TMAO shared 20 pathways of these 118 CRCrelated pathways with CRC. The top 10 ranked common pathways between TMAO and CRC (OMIM), TMAO and CRC(GWAS), and TMAO and CRC-genes from both OMIM and GWAS are shown in Table 6.