RecallPrecision {MRPC}

Calculate recall and precision for two graphs

Description

This function counts the number of true and false positives, and calculates recall (i.e., power) and precision (i.e., 1-FDR), which are defined as follows:

Recall = (# edges correctly identified in inferred graph) / (# edges in true graph).

Precision = (# edges correctly identified in inferred graph) / (# edges in inferred graph).

Usage

RecallPrecision(g1, g2, GV, includeGV, edge.presence = 1.0, edge.direction = 0.5)

Arguments

g1

First graph object, from the true graph

g2

Second graph object, from the inferred graph

GV

The number of genetic variants (SNPs/indels/CNV/eQTLs) in the input data matrix. For example, if the data has one variant, which is in the first column, then GV = 1. If there are two variants, which are in the first and second Columns, then GV = 2. If there are no variants, then GV = 0.

includeGV

If TRUE, include edges involving genetic variants (GV) when calculating recall and precision. If FALSE, exclude edges involving genetic variants (GV) when calculating recall and precision.

edge.presence

The weight for an edge being present.

edge.direction

The weight for the edge direction.

Details

We consider it more important to be able to identify the presence of an edge than to also get the direct correct. Therefore, we assign 1 as the default to an edge with the correct direction and 0.5 to an edge with the wrong direction or no direction (Badsha and Fu, 2019; Badsha et al., 2021).

Value

A list of object that containing the following:

• Matrix: Results store for TP and FP

• TP: Total found edges in the inferred graph and edge exists in the true graph.

• FP: Total found edges in the inferred graph but no edge exists in the true graph.

• NTE: Total number of edges in the true graph.

• NIE: Total number of edges in the inferred graph.

• Recall: Power, or sensitivity measures how many edges from the true graph a method can recover.

• Precision: Measures how many correct edges are recovered in the inferred graph.

Author(s)

Md Bahadur Badsha ([e-mail])

References

1. Badsha MB and Fu AQ (2019). Learning causal biological networks with the principle of Mendelian randomization. Frontiers in Genetics, 10:460.

2. Badsha MB, Martin EA and Fu AQ (2021). MRPC: An R package for inference of causal graphs. Frontiers in Genetics, 10:651812.

See Also

aSHD: adjusted Structural Hamming Distance (aSHD)

Examples

# True model
# True graph (V1 --> T1 --> T2 --> T3) 
# Where V1 is a genetic variant (GV) and T1, T2, and T3 are phenotypes
tarmat_s1 <- matrix(0,
                    nrow = 4,
                    ncol = 4)
                    
colnames(tarmat_s1) <- c("V1", "T1", "T2", "T3")
rownames(tarmat_s1) <- colnames(tarmat_s1)

# Create an adjacency matrix for the true graph
tarmat_s1[1, 2] <- 1
tarmat_s1[2, 3] <- 1
tarmat_s1[3, 4] <- 1

# Graph object of the true graph
Truth <- as(tarmat_s1,
            "graphNEL")

# Inferred graph (V1 --> T1 <-- T2 --> T3)
# Where V1 is a genetic variant (GV) and T1, T2, and T3 are phenotypes
tarmat_s2 <- matrix(0,
                    nrow = 4,
                    ncol = 4)
                    
colnames(tarmat_s2) <- c("V1", "T1", "T2", "T3")
rownames(tarmat_s2) <- colnames(tarmat_s2)

# Create an adjacency matrix for the inferred graph
tarmat_s2[1, 2] <- 1
tarmat_s2[3, 2] <- 1
tarmat_s2[3, 4] <- 1

# Graph objects for the inferred graph
Inferred <- as(tarmat_s2,
               "graphNEL")
         
# Recall and Precision
Recall_Precision <- RecallPrecision(Truth,
                                   Inferred,
                                   GV = 1,
                                   includeGV = TRUE,
                                   edge.presence = 1.0,
                                   edge.direction = 0.5)