An atlas of evidence-based phenotypic associations across the mouse phenome

To date, reliable relationships between mammalian phenotypes, based on diagnostic test measurements, have not been reported on a large scale. The purpose of this study was to present a large mouse phenotype-phenotype relationships dataset as a reference resource, alongside detailed evaluation of the resource. We used bias-minimized comprehensive mouse phenotype data and applied association rule mining to a dataset consisting of only binary (normal and abnormal phenotypes) data to determine relationships among phenotypes. We present 3,686 evidence-based significant associations, comprising 345 phenotypes covering 60 biological systems (functions), and evaluate their characteristics in detail. To evaluate the relationships, we defined a set of phenotype-phenotype association pairs (PPAPs) as a module of phenotypic expression for each of the 345 phenotypes. By analyzing each PPAP, we identified phenotype sub-networks consisting of the largest numbers of phenotypes and distinct biological systems. Furthermore, using hierarchical clustering based on phenotype similarities among the 345 PPAPs, we identified seven community types within a putative phenome-wide association network. Moreover, to promote leverage of these data, we developed and published web-application tools. These mouse phenome-wide phenotype-phenotype association data reveal general principles of relationships among mammalian phenotypes and provide a reference resource for biomedical analyses.

and are displayed in a tree view according to 60 biological systems (above). When a phenotype of interest (in this case, 'adult_trait:abnormal adipose tissue amount') is selected from the above tree, associations between the selected phenotype and its related phenotypes (PPAP or phenotypic expression module) are interactively displayed (below). When the mouse pointer is positioned over the query phenotype in the middle of the PPAP, the number of incoming phenotypes (indegree), the number of outgoing phenotypes (outdegree), the sum of indegree and outdegree (i.e., degree), and the number of distinct biological systems for phenotypes belonging to the displayed PPAP are designed to appear during the mouse over. By contrast, when the mouse pointer is      ・ Support: Co-expression frequency of the two abnormal phenotypes constituting each association rule.
・ Confidence: In an association rule between phenotypes, when the phenotype on the left-hand side (premise) is abnormal, the probability that the phenotype on the right-hand side (conclusion) is abnormal. Generally, this measure represents the strength of the rule.
・ Rule polarity: A measure representing the magnitude of the difference between the two confidence values in a bidirectional rule. Considering the formula for calculating the confidence value, the larger the value of this measure, the larger the difference between the number of abnormal cases in the premise and those in the conclusion of an association rule. Therefore, this measure indicates the relative degree of difference between the number of abnormal cases in the premise and those in the conclusion of an association rule. In this study, because an association rule that had a larger confidence value in a bidirectional rule was selected as one of the 3,686 significant rules, a larger value of this measure represents a relatively smaller number of abnormal cases in the premise.
・ Rule significance: A measure demonstrating the statistical significance of an association rule between phenotypes. The larger the value of this measure, the more reliable the association rule.
We performed various analyses with the following objectives:

How are values of the four measures distributed?
Refer to Supplementary Fig. 2a-d. The mean support value was 2.4%, and the median was 1% (Supplementary   Fig. 2a). The mean confidence value was 48.2%, the median was 42.9%, and the mode was 18% (Supplementary   Fig. 2b). For rule polarity values, the mean was 0.58 and the median was 0.32 (Supplementary Fig. 2c). For rule significance values, the mean was 4.25 and the median was 2.86 (Supplementary Fig. 2d). Characteristics of the distributions of values of these four measures by stage/type and biological system rule categories are shown in Supplementary Fig. 2f-j and SD 13 Figs. 1-5, respectively.

How are values of the four measures correlated?
Refer to Supplementary Fig. 2e. Spearman's rank correlation coefficients were calculated to evaluate correlations between the four measures described above (Supplementary Fig. 2e). There were strong positive correlations between three pairs of measures: support and confidence, support and rule significance, confidence and rule significance (r = 0.79, 0.52, 0.44, respectively). The probabilities of type I error were practically nil for all the three pairwise comparisons. These results indicate interdependent relationships among these three measures.

Which 'stage/type' rule categories are enriched among the upper and lower values for each of the four measures?
Refer to Supplementary

What are the distributions of values of the four measures for each of the eight 'stage/type' rule categories constituting the 3,686 significant rules?
Refer to Supplementary Fig. 2f-i. Supplementary Fig. 2f-i Fig. 2j). In the comparison 'adult_trait => adult_trait: adult_gene => adult_gene', which covered 92.9% (3,424) of all the significant rules (3,686), we found remarkably significant differences for all four measures (P = 0, 0, 5.4×10 -11 , and 2.2×10 -10 , respectively). This result indicates that, in comparisons of rules between trait expressions and rules between gene expressions at the adult stage, there are marked differences in the values of all four measures. Also, significant differences were observed in all four measures at 80% or more for all ten comparisons between 'stage/type' rule categories (at the 0.05 significance level). Therefore, the following analyses of the four measures for extracting features of rules between biological systems were performed according to the five types of between-'stage/type' rule categories.

According to between-'biological systems' rules of the between-'stage/type' rule 'adult_trait => adult_trait'
Refer to Supplementary Table 9 and SD 13 Fig. 1. The between-'stage/type' rule 'adult_trait => adult_trait' consisted of 244 distinct between-'biological systems' rules ( Fig. 2d-i), of which 86 had seven or more 'between-phenotypes' rules. We analyzed the 86 rules to extract features of the values of the four measures (support, confidence, rule polarity, rule significance) in 'between-phenotypes' rules (SD 13 Fig. 1a-c, 1d-f, 1g-i, 1j- The above four between-'biological systems' rules comprised only two kinds of biological system, 'adult_trait:hematopoietic system phenotype(MP)' and 'adult_trait:immune system phenotype(MP)', and exhibited much higher mean values for support, confidence, rule polarity, and rule significance for 'between-phenotypes' rules belonging to each of the four between-'biological systems' rules (SD 13 Fig. 1c, f, i,

Rules between biological systems Label (enrichment rank)
Num. of rules between phenotypes above four between-'biological systems' rules exhibited greater co-expression frequencies in rules comprising two phenotypes, greater strength in 'between-phenotypes' rules, greater difference in the number of abnormal cases between two phenotypes constituting a rule, and greater statistical significance for 'between-phenotypes' rules.
Also, because all ontology-annotated phenotypes based on parameters measured by the immunophenotyping (FACS) test were classified into the above two biological systems ('adult_trait:hematopoietic system phenotype(MP)' and 'adult_trait:immune system phenotype(MP)'; refer to Supplementary Table 1), almost all of the rules between abnormal phenotypes belonging to each of the above four between-'biological systems' rules were, in fact, derived from relationships among the parameters measured in the FACS test. In addition, note that 'between-phenotypes' rules belonging to each of the above four between-'biological systems' rules were positively selected with statistical significance during the extraction of the 3,686 significant rules (refer to Fig. 2d and Supplementary Table 6).

・ Example 2: Characteristic rules between biological systems in the 'adult_trait => adult_trait' category
The above three between-'biological systems' rules were identified as exhibiting lower mean values for confidence in 'between-phenotypes' rules belonging to each of the three between-'biological systems' rules (18.7%, 17.4%, and 19.6%, respectively; SD 13 Fig. 1f). That is, we detected remarkable features, indicating that the rules between abnormal phenotypes belonging to each of the above three between-'biological systems' rules exhibited lower values for the strength of rules between phenotypes. In addition, the between-'biological systems' rule, label 85, exhibited extremely low mean values for support (0.2%) in rules between phenotypes belonging to the label 85 rule, enabling us to detect a low-frequency between-'biological systems' rule, which would be difficult to find as a significant relationship without using the rule selection criteria and data from comprehensive phenotyping analyses, both of which were applied in this study. Further, note that 'between-phenotypes' rules belonging to each of the above three between-'biological systems' rules were positively selected with statistical significance during extraction of the 3,686 significant rules (refer to Fig. 2d and Supplementary Table 6).

・ Example 3: A characteristic rule between biological systems in the 'adult_trait => adult_trait' category
The above between-'biological systems' rule was identified as exhibiting the highest mean value for rule polarity (0.8; SD 13 Fig. 1i) among 'between-phenotypes' rules belonging to this between-'biological systems' rule. That is, we detected a feature that, in 'between-phenotypes' rules belonging to this between-'biological systems' rule, the relative difference in the number of abnormal cases for each of the 'between-phenotypes' rules was remarkably large. This feature indicates that abnormal phenotypes belonging to the biological system 'adult_trait:homeostasis/metabolism phenotype(MP)' exhibit a relatively small number of abnormal phenotypic expression cases, while abnormal phenotypes in the biological system 'adult_trait:growth/size/body region

Rules between biological systems Label (enrichment rank)
Num. of rules between phenotypes phenotype(MP)' exhibit a relatively large number of cases of abnormal phenotypic expression. Note that 'between-phenotypes' rules belonging to this between-'biological systems' rule were positively selected with statistical significance during extraction of the 3,686 significant rules ( Fig. 2d and Supplementary Table 6). ・ Example 4: Characteristic rules between biological systems in the 'adult_trait => adult_trait' category The above four between-'biological systems' rules were identified as exhibiting extremely small mean values of rule polarity for 'between-phenotypes' rules belonging to each of these four between-'biological systems' rules (SD 13 Fig. 1i). That is, we detected the feature that, in 'between-phenotypes' rules belonging to each of these four between-'biological systems' rules, the relative difference in the number of abnormal cases for each 'between-phenotypes' rule was remarkably small. This feature demonstrates that, in these four between-'biological systems' rules, there are a large number of 'between-phenotypes' rules, each of which exhibits a similar number of abnormal cases in the two abnormal phenotypes constituting the rule. In addition, note that 'between-phenotypes' rules for the between-'biological systems' rule 'adult_trait:homeostasis/metabolism phenotype(MP) => adult_trait:homeostasis/metabolism phenotype(MP)' (label 8), in which abnormal phenotypes were mainly derived from parameters measured in clinical blood chemistry tests, were positively selected with statistical significance during extraction of the 3,686 significant rules ( Fig. 2d and Supplementary Table 6).

・ Example 5: A characteristic rule between biological systems in the 'adult_trait => adult_trait' category
The above between-'biological systems' rule was identified as exhibiting the highest mean value for the measure significance (4.7, SD 13 Fig. 1l) among 'between-phenotypes' rules belonging to this between-'biological systems' rule. That is, we detected the feature that 'between-phenotypes' rules belonging to this between-'biological systems' rule exhibited higher significance. In addition to this between-'biological systems' rule, the four between-'biological systems' rules presented in example 1 have the same feature. Further, note that 'between-phenotypes' rules belonging to this between-'biological systems' rule were mostly positively selected with statistical significance during extraction of the 3,686 significant rules ( Fig. 2d and Supplementary Table 6).
For further details of the mean values for each of the four measures for between-'biological systems' rules belonging to the between-'stage/type' rule 'adult_trait => adult_trait', refer to the 'heatmap' sheets for each measure in Supplementary Table 9.  Fig. 3. In this study, all of the phenotypes annotated with the phenotypic category 'adult_gene' were derived from the parameters measured by the adult LacZ test. Therefore, note that between-'biological systems' rules belonging to the between-'stage/type' rule 'adult_gene => adult_gene' correspond to rules between MA ontology-annotated tissues, where abnormal gene expression was observed. The between-'stage/type' rule 'adult_gene => adult_gene' consists of 200 distinct between-'biological systems' rules ( Fig. 2d-i), among which, 56 had seven or more 'between-phenotypes' rules. Here, we analyzed these 56 rules to extract features of the values of the four measures (support, confidence, rule polarity, rule significance) of rules between phenotypes (SD 13 Fig. 2a-c, 2d-f, 2g-i, and 2j-l, respectively). For each of the four measures, the results of basic statistical analyses, Levine's test for equality of variance, ANOVA, post-hoc test (Games-Howell method), and hierarchical clustering using the P values resulting from the post-hoc test are detailed in Supplementary Table 10. By hierarchical clustering using P values from these post-hoc tests, we specified between-'biological systems' rules with remarkably high/low mean values for support, confidence, rule polarity, and rule significance (SD 13 Fig. 2c, f, i, and l, respectively). In the between-'stage/type' rule 'adult_gene => adult_gene', we finally identified 30 distinct characteristic rules between biological systems, which are summarized in the table in the lower part of Fig. 3b. Below, we present representative examples from the 30 distinct rules between biological systems. ・ Example 1: A characteristic rule between biological systems in the 'adult_gene => adult_gene' category

According to between-'biological systems' rules belonging to the between-'stage/type' rule 'adult_gene => adult_gene'
The above between-'biological systems' rule, comprising a single biological system, 'adult_gene:nervous system(MA)', exhibited extremely high mean values for support (8.7%, SD 13 Fig. 2c), confidence (89%, SD 13 Fig. 2f), and rule significance (12, SD 13 Fig. 2l), among 'between-phenotypes' rules belonging to this between-'biological systems' rule. That is, we detected remarkable features, indicating that the between-'abnormal phenotypes' rules belonging to this between-'biological systems' rule exhibited a greater co-expression frequency in two phenotypes constituting rules, greater strength in 'between-phenotypes' rules, and greater statistical significance in 'between-phenotypes' rules. In addition, note that 'between-phenotypes' rules belonging to this between-'biological systems' rule were positively selected with statistical significance during extraction of the 3,686 significant rules ( Fig. 2d and Supplementary Table 6). ・ Example 2: A characteristic rule between biological systems in the 'adult_gene => adult_gene' category The above between-'biological systems' rule exhibited an extremely low mean value for support (2.3%, SD 13 Fig.   2c) while having extremely high mean values for both confidence (85%, SD 13 Fig. 2f) and polarity (1.5, SD 13 Fig. 2i), for 'between-phenotypes' rules belonging to this between-'biological systems' rule. That is, we detected remarkable features, indicating that between-'abnormal phenotypes' rules belonging to this between-'biological systems' rule exhibited lower co-expression frequency in two phenotypes constituting rules, while exhibiting both greater strength of 'between-phenotypes' rules and greater difference in the number of abnormal cases in each 'between-phenotypes' rule. In addition, note that 'between-phenotypes' rules belonging to this between-'biological systems' rule were positively selected with statistical significance during extraction of the 3,686 significant rules ( Fig. 2d and Supplementary Table 6). ・ Example 3: A characteristic rule between biological systems in the 'adult_gene => adult_gene' category The above between-'biological systems' rule exhibited extremely low mean values for both support (1.4%, SD 13  Fig. 2i), for 'between-phenotypes' rules belonging to this between-'biological systems' rule. That is, we detected remarkable features, indicating that between-'abnormal phenotypes' rules belonging to this between-'biological systems' rule exhibited both lower co-expression frequencies in the two phenotypes constituting rules and lower statistical significance for 'between-phenotypes' rules, while exhibiting both greater strength of 'between-phenotypes' rules and a greater difference in the number of abnormal cases in each 'between-phenotypes' rule. In addition, note that 'between-phenotypes' rules belonging to this between-'biological systems' rule were positively selected with statistical significance during extraction of the 3,686 significant rules ( Fig. 2d and Supplementary Table 6).
For further details of the mean values for each of the four measures of between-'biological systems' rules belonging to the between-'stage/type' rule 'adult_gene => adult_gene', refer to the 'heatmap' sheets for each measure in Supplementary Table 10.

According to between-'biological systems' rules belonging to the between-'stage/type' rule 'adult_trait => adult_gene'
Refer to Supplementary Table 11 and SD 13 Fig. 3. The between-'stage/type' rule 'adult_trait => adult_gene' comprised 112 'between-phenotypes' rules covering 69 distinct between-'biological systems' rules (right table, Fig.   2d-i). Of the 69 between-'biological systems' rules, there were only three rules that had seven or more 'between-phenotypes' rules (table below). Therefore, in this study we present only distributions of values in the four measures (support, confidence, rule polarity, rule significance) for each 'between-phenotypes' rule (SD 13 Fig.   3a-d, respectively) and their fundamental statistics (Supplementary Table 11), according to the 69 distinct between-'biological systems' rules. Furthermore, note that 'between-phenotypes' rules belonging to each of the three above-described between-'biological systems' rules were not positively/negatively selected with statistical significance during extraction of the 3,686 significant rules (refer to Fig. 2d and

According to the between-'biological systems' rules belonging to the between-'stage/type' rule 'adult_gene => adult_trait'
Refer to Supplementary Table 13 and SD 13 Fig. 5. The between-'stage/type' rule 'adult_gene => adult_trait' comprised 55 'between-phenotypes' rules covering 42 distinct between-'biological systems' rules (right table, Fig.   2d-i). Of the 42 between-'biological systems' rules, only two had seven or more 'between-phenotypes' rules (refer to the table below). Therefore, in this study we present only distributions of values for the four measures (support, confidence, rule polarity, rule significance) for each 'between-phenotypes' rule (SD 13 Fig. 5a-d, respectively), and their fundamental statistics (Supplementary Table 13), according to the 42 distinct between-'biological systems' rules. Furthermore, note that 'between-phenotypes' rules belonging to each of the two above-described between-'biological systems' rules were negatively selected with statistical significance during extraction of the 3,686 significant rules (refer to Fig. 2d and Fig. 1 Identifying between-'biological systems' rules exhibiting marked values for each of the four measures (support, confidence, rule polarity, rule significance) for 'between-phenotypes' rules for the 'adult_trait => adult_trait' between-'stage/type' rule. Of 244 kinds of between-'biological systems' rules in this between-'stage/type' rule, 86 where each rule category had a sample size ≥ 7 were applied for analysis. The between-'biological systems' rules examined were labeled according to rankings from the results of rule enrichment analysis for 608 kinds of between-'biological systems' rules (Supplementary Table 6). Results of analysis of values of each of the four measures (support, confidence, rule polarity, rule significance) for the 'between-phenotypes' rules are shown in SD 13 Fig. 1a-c, SD 13 Fig. 1d-f, SD 13 Fig. 1g-i, and SD 13 Fig. 1j-l,   respectively. (a,d,g,j) Distributions of values for each of the four measures by between-'biological systems' rules.
In these panels, for each of four measures, the 86 between-'biological systems' rules examined are arranged along the x-axis, in descending order (from the left) of mean value. Asterisks in red and blue represent between-'biological systems' rules with much larger/smaller values in the measure of interest, respectively.
(b,e,h,k) Clustered heatmaps with dendrograms exhibiting mean differences in each measure among between-'biological systems' rules. By hierarchical clustering (Euclidean distance and Ward's linkage) of between-'biological systems' rules, using the P values resulting from the post-hoc test (Games-Howell method) after one-way ANOVAs, between-'biological systems' rules with remarkably high/low mean values in each of support, confidence, rule polarity, and rule significance were identified. Note that the heatmap for each measure is symmetrical. Colored squares on heatmaps represent relationships with post-hoc P < 0.05, and darker colors represent greater differences between between-'biological systems' rules. For each dendrogram, sub-clusters in red and blue indicate between-'biological systems' rules with relatively greater and smaller values, respectively.   Table 6). Results of analysis of values for each of the four measures (support, confidence, rule polarity, rule significance) for the 'between-phenotypes' rules are shown in SD 13 Fig. 2a-c, 2d-f, 2g-i, and 2j-l, respectively. (a,d,g,j) Distributions of values for each of the four measures, according to between-'biological systems' rules. In these panels, the 86 between-'biological systems' rules   . 6); therefore, cluster 4 could be regarded as a sub-cluster of this large cluster.
• Refer to the table below. Cluster 4 consisted of phenotypes covering 25 biological systems, where only a set of phenotypes belonging to the biological system 'adult_trait:vision/eye phenotype(MP)' predominantly contributed to the statistical significance (corrected P = 9.2×10 -3 ).

Enrichment analysis of phenotypes in cluster 4 according to their biological systems
• As a characteristic example of the PPAPs in this cluster, we present a PPAP with the query phenotype 'adult_trait:abnormal coping response', which was the only phenotype belonging to the biological system 'adult_trait:behavior/neurological phenotype(MP)' in this cluster (see the figure below).
Enrichment analysis of phenotypes in cluster 5 according to their biological systems ***corrected P < 0.001.
An example PPAP in cluster 5. A PPAP with the query phenotype 'adult_trait:abnormal coping response' is shown.
All six phenotypes related to the query phenotype belonged to both 'adult_trait:immune system phenotype(MP)' and 'adult_trait:hematopoietic system phenotype(MP)'. These six phenotypes are related to abnormal cell numbers in the T cell lineage. As edge arrows and edge color gradations represent the direction and strength of rule polarity, respectively, all six phenotypes, particularly 'adult_trait:abnormal T cell number' and 'adult_trait:abnormal CD8-positive, alpha-beta T cell number', exhibit larger numbers of abnormal cases compared with the query phenotype.

Cluster 6
• Cluster 6 consisted of 33 phenotypes (Supplementary Table 2 remaining four phenotypes belonged to the 'embryo/trait' category, and phenotypes in this 'stage/type' category were significantly enriched in the neighboring cluster, cluster 6. • Refer to the table below. Cluster 7 consisted of phenotypes covering 11 biological systems, where phenotypes belonging to the biological systems 'adult_trait:immune system phenotype(MP)' and 'adult_trait:hematopoietic system phenotype(MP)' were predominantly responsible for the statistical significance (corrected P = 1.5×10 -3 , 2.5×10 -3 , respectively). Note that the feature of cluster 7, that phenotypes belonging to each of the above two biological systems were significantly enriched, was the same as that observed in cluster 5.
• Of the 36 phenotypes in this cluster, 24 belonged to the biological systems 'adult_trait:immune system phenotype(MP)' and 'adult_trait:hematopoietic system phenotype(MP)'. Among these 24 phenotypes, those participating in the innate immune system, such as 'adult_trait:abnormal monocyte cell number', 'adult_trait:abnormal dendritic cell number', and 'adult_trait:abnormal neutrophil cell number', and B cell immunity-related phenotypes participating in the acquired immune system, such as 'adult_trait:abnormal B cell number', 'adult_trait:abnormal immature B cell number', and 'adult_trait:abnormal mature B cell number' were particularly prominent.
• As a characteristic example of the PPAPs in this cluster, we present a PPAP with the query phenotype 'adult_trait:abnormal B cell number', which had the largest number of constituting phenotypes (see the figure below). Ratio in total ** ** Supplementary Data 9 was then built by deleting records where the number of individuals in the control group was ≤ 1 or the number of individuals in the mutant group was 0 from this table.

Creating a table of phenotypic calls for fertility and viability
Phenotypic calls for two kinds of tests, fertility and viability, linked to each experiment group (colony_id), not each individual, are summarized in Supplementary Data 10 and Supplementary Data 11, respectively. The results of manual calls of phenotypes (normal or abnormal) for each experiment group (colony_id) are presented in the column 'call_fertility_fecundity' in Supplementary Data 10 and in the column 'call_viability_by_preweaning' in Supplementary Data 11. These two types of data were used for subsequent association rule mining as the two phenotypes 'adult_trait:abnormal fertility/fecundity' and 'embryo_trait:abnormal viability by preweaning', respectively.

Integration of semantically identical phenotypes
To integrate semantically identical phenotyping results, we reshaped the data in Supplementary Data 9, according to genes (gene symbols), based on the value (phenotype) in the column 'stage_type_ontology name'. Specifically, to make the reshaped dataset (Supplementary Data 12), when multiple measured parameters were identified in the same phenotype (from the column 'stage_type_ontology_name') in each mutant gene categorization (from the column 'gene_symbol'), the one with the highest degree of abnormality was selected as representative of the phenotype, in priority order of the one with the smallest P value for ES (unbiased Hedge's g), the one with the smallest P value for the significance test, and the one with the largest ES.

A dataset for association rule mining
A dataset obtained by adding phenotypic calls from Supplementary Data 10 (Fertility) and Supplementary Data 11 (Viability) into the table, Supplementary Data 12, was used to construct a normal/abnormal phenotypic call matrix of 532 phenotypes × 3,100 gene symbols (mutant strains) for subsequent association rule mining. The 532 phenotypes were derived from 2,050 measured parameters, and cover 84 distinct biological systems (top level terms), classified into four 'stage/type' phenotype expression categories: 35 in 'adult/trait', 16 in 'adult/gene', 16 in 'embryo/trait', and 17 in 'embryo/gene' (Supplementary Table 1).