BEERE is the abbreviation of Biomedical Entity Expansion, Ranking, and Explorations, which can help biomedical researchers investigate the relevant significance of genes, proteins, and general biomedical concepts—biomedical entities—among one another from public knowledge-base of protein/gene interactions and extracted semantic relationships. BEERE aims to assist users to quickly evaluate and prioritize a list of genes or terms, i.e., “entities”, based on our established network ranking technique (1) which can take advantage of a network of probabilistic functional relationships extracted a priori. It also provide the significance of genes in functional gene set or networks.


The input to BEERE can be either a list of gene names, which biomedical users obtain from functional genomic analysis using next-generation sequencing data, or a list of generic biomedical terms, which biomedical users obtain from hypothesis-driven research.
The output from BEERE has three parts: 1) A new list of gene or term entities from input entities expanded from the a priori knowledge base, using the input list of entities as seeds; 2) New rank information of all the input seed entities and the extended list of entities with added entity relationship network ranking information, including its rank, a ranking score, and adjusted p-value indicating the statistical significance of the ranking score in the newly formed entity relationship network against randomly permutated networks; and 3) A graphical display of the newly formed entity relationship network, drawn based on user preferences of circular, force-directed, and DEMA layouts, with the ability to help biomedical users drill down on specific relationship(s) of interest for database source entry or source publication. Users of BEERE will also find details on the network of probabilistic functional relationships among terms extracted, summary statistics consisting of rank score distribution and word-cloud summary of all terms with size drawn according to their ranking scores. BEERE provides biomedical users to export the output data, including both tables and networks, as tabular or image files. To provide users with new rank information.


The SEMANTIC types can be mapped to the full names using the parsable list of Semantic Types and their abbreviations from the UMLS via the link. https://metamap.nlm.nih.gov/Docs/SemanticTypes_2018AB.txt


The input parameters are:

"biomedical entity": a list of biomedical entities separated by enter;
"extended": "yes" or "no" for the network one-layer extension;
"ppi" (gene Symbol): the Protein-Protein Interaction confidant score cutoff (defalt = 0.45, range from 0 to 1);
"-logPMF" (biomedical term): the -logPMF cutoff of hypergeometric test using the predication counts (defalt = 0);
"fuzzy matched" (biomedical term only): do a fuzzy matching to the terms in SEMMED database;
"iteration": the iteration for the recursive RP-score (default = 200, range from 0 to 1,000);
"sigma": the damping factor (default = 0.8, range from 0 to 1).
"method": "Ant colony", "PageRank" (default = "Ant colony").

Features Description
Biomedical entity prioritization Rank entity in the network or the one-layer extended network consisting of the input entity list.
Word-cloud visualization of the entity ranking score (RP-score) using word-cloud.
P-value estimation Ranking of the RP-score and take the 1% of RP-score from two-tail and mark as very significant (**), and the 5% of RP-score from two-tail and mark as significant (*).
API calling The BEERE’s api is designed for data analysis using language R or python. The connect method could be both “Get” and “Post” from the site: http://discovery.informatics.uab.edu/BEERE/index.php/search/api/.


Entity prioritization:
BEERE provides two types of ranking algorithms to prioritize biomedical entities. For both of the algorithm, we initiate the ranking score applying the heuristic scoring function (Initial Ranking Algorithm) to calculate the initial ranking score:
RP(i;0) = e(k × ln(∑∃(i,j)RC(i,j)) - ln(∑∃(i,j)N(i,j))),
where i and j are the indexes of proteins from the selected module, k is a constant (k=2 in this study). The relation confidence score RC(i, j) can be either assigned by the protein interaction confidence in HAPPI-2 database or the RDS calculated using semantic predications in SemMedDB.
Two iterative scoring algorithms have been provided.
In the page rank algorithm, we perform iterative scoring and ranking to calculate the page ranking score (PR) by the formula:
PR(i;t) = (1-σ) × PR(i;0) + σ × (∑∃(i,j)conf(i,j) × PR(j;t-1))/(∑∃(i,j)N(i,j)),
Where t is the number of iterations (default = 200), σ is the damping factor (default = 0.8).
In the ant-colony algorithm, the ant-colony ranking score (AR) calculated using the inflow information and outflow information in the formula:
AR(i;t) = AR(i;t-1) - σ × Δ(i,j) × AR(i;t-1) + σ × Δ(j,i) × AR(j;t-1),
where t is the number of iterations (default = 200),σ is the damping factor (default = 0.8) representing the probability of a node continuing the ‘information flow’. Δ((i,j) and Δ((j,i) represent the normalized information using:
Δ(i,j) = ∑(i,j)RC(i,j)/∑l∈(i,l)RC(i,l),
Δ(j,i) = ∑(i,j)RC(j,i)/∑k∈(k,j)RC(k,j),
P-value estimation:
We rank the RP-score and take the 1% of RP-score from two-tail and mark as very significant (**), then we take the 5% of RP-score from two-tail and mark as significant (*).


The BEERE’s APIs provide programmatic access to various services that deal analysis using language R, python and etc. Users can use method “Post” to query the result using the parameters listed below:

API detail will be added later

Edge prioritization:
Firstly, we perform path weighted matrix (PWM) construction using every node-to-node highest weighted pathway by applying the dijkstra likewise algorithm (multiple the PPI socre).
Secondly, we perform the edge weighted matrix (EWM) construction using the path weighted matrix score in formula:
EWM(E(Ai,Aj) - E(Bi,Bj)) = conf(Ai,Aj)×conf(Bi,Bj) × max⁡(PWM(A(i)-B(i)),PWM(A(j)-B(i)),PWM(A(j)-B(i)),PWM(A(j)-B(j))),
Where conf(Ai,Aj) is the PPI confidant score of edge A consisting of node i and node j.
Thirdly, we drop off the bottom 25% values from EWM and feed remain values to the iterative scoring and ranking formula same above to calculate the edge score.
Browser Compatibility
OS Version Chrome Firefox Microsoft Edge Safari
Windows 10 70.0.3538.77 64.0 17.17134 n/a
MacOS HighSierra 70.0.3538.77 64.0 17.17134 5.1