drugfindR

Pipeline Components

The five pipeline functions are:

1. get_signature(): This function takes a LINCS ID and returns the corresponding signature.
2. prepare_signature(): This function takes a transcriptomic signature and prepares it for analysis by drugfindR.
3. filter_signature(): This function takes a signature and filters it to given thresholds.
4. get_concordants(): This function takes a signature and returns the concordant signatures from the iLINCS database.
5. consensus_concordants(): This function takes a list of concordant signatures and returns a list of consensus signature.

Step 1: Get the Signature

This signature is available from Zenodo here. Our first step is to download the signature so we can work with it. The read_tsv() function from the readr package can be used to read the signature into R from a remote URL.

# Load the signature from the paper

diffexp <- read_tsv("https://zenodo.org/records/10265182/files/dCovid_diffexp.tsv")
#> Rows: 17117 Columns: 6
#> ── Column specification ────────────────────────────────────────────────────────
#> Delimiter: "\t"
#> chr (1): hgnc_symbol
#> dbl (5): logFC, logCPM, F, PValue, FDR
#> 
#> ℹ Use `spec()` to retrieve the full column specification for this data.
#> ℹ Specify the column types or set `show_col_types = FALSE` to quiet this message.

# Take a look at the signature

head(diffexp) |>
    knitr::kable()

We can see that the signature has ncol(diffexp) columns and nrow(diffexp) rows. The names of the columns are typical of what you would get from edgeR or DESeq2.

Step 2: Prepare the Signature

The next step is to prepare the signature for analysis by drugfindR. This step is necessary because the signature can be in many different formats, with different names for columns. iLICNS needs columns to be in a specific order and with specific names. The prepare_signature() function takes care of this for us.

prepare_signature() takes three optional arguments:

1. gene_column: The name of the column in the input that contains the gene names. The default is "Symbol".
2. logfc_column: The name of the column in the input that contains the log fold change values. The default is "logFC".
3. pval_column: The name of the column in the input that contains the p-values. The default is "PValue".

# Prepare the signature for analysis
# The only thing that is different from the defaults is the gene_column
# However, we will specify all three arguments for clarity

signature <- prepare_signature(diffexp, gene_column = "hgnc_symbol", logfc_column = "logFC", pval_column = "PValue")

# Take a look at the signature

head(signature) |>
    knitr::kable()

We can see that the signature has been reordered and renamed. The first column is now names(signature)[1], the second column is now names(signature)[2], and the third column is now names(signature)[3], which is what iLINCS expects.

Step 3: Filter the Signature

Now that we have the signature in the correct format and filtered to the L1000 genes, we can filter it to the thresholds that we want. This filter step is necessary because we would like to use the genes that have a high enough change for it to matter.

The filter_signature() function can filter based on logFC values in two ways:

1. Absolute Threshold: You can give an absolute threshold (or a pair of absolute thresholds) for the logFC values. Any genes that do not meet the threshold will be removed from the signature.
2. Percentile Threshold: You can give a percentile threshold (or a pair of percentile thresholds) for the logFC values. Any genes that do not meet the threshold will be removed from the signature.

The filter_signature() function takes three arguments:

1. signature: The signature to filter.
2. direction: This argument specifies whether to filter for upregulated genes, downregulated genes, or both. The default is "any".
3. One of threshold or prop: The threshold argument is used to specify an absolute threshold (or a pair of absolute thresholds) for the logFC values. The prop argument is used to specify a percentile threshold (or a pair of percentile thresholds) for the logFC values. They can not be specified together.

# Filter the signature to only include genes that are upregulated by at least 1.5 logFC

filtered_signature_up <- filter_signature(signature, direction = "up", threshold = 1.5)

filtered_signature_up |>
    head() |>
    knitr::kable()

# Filter the signature to only include genes that are downregulated by at least 1.5 logFC
filtered_signature_dn <- filter_signature(signature, direction = "down", threshold = 1.5)

filtered_signature_dn |>
    head() |>
    knitr::kable()

Step 4: Get the Concordant Signatures

Now that we have the filtered signatures for both upregulated and downregulated genes, we can get the concordant signatures from the iLINCS database. The get_concordants() function takes a signature and returns the concordant signatures from the iLINCS database. It also requires specification of the database to target for the concordant signatures.

The get_concordants() function takes the following arguments:

1. signature: The signature to get concordant signatures for.
2. ilincs_library: The iLINCS library to target for concordant signatures. This can be one of c(“OE”, “KD”, “CP”), standing for overexpression, knockdown, and chemical perturbagens, respectively.
3. direction: This argument specifies whether the input signature is upregulated or downregulated. This is useful to annotate the output. This is NULL by default.

# Get the concordant signatures for the upregulated signature

up_concordants <- get_concordants(filtered_signature_up, ilincs_library = "CP", sig_direction = "up")

up_concordants |>
    head() |>
    knitr::kable()

# Get the concordant signatures for the downregulated signature

dn_concordants <- get_concordants(filtered_signature_dn, ilincs_library = "CP", sig_direction = "down")

dn_concordants |>
    head() |>
    knitr::kable()

Step 5: Get the list of Consensus Concordant Signatures

Now that we have the concordant signatures for both the upregulated and downregulated signatures, we can get the list of consensus concordant signatures. The consensus_concordants() function takes a list of concordant signatures and returns a list of consensus signatures. This function also takes a number of optional arguments that can be used to control the consensus list generation.

By default the consensus list performs the following steps:

1. Combine the list of concordant signatures into a single data frame.
2. For each individual signature origin (Gene or Drug), choose the one with the largest absolute concordance value.

Additionally we can filter by the cell line to only include the cell lines of interest.

The consensus_concordants() function takes the following arguments:

1. ...: One or Two (see paired) Data Frames with the concordants
2. paired: A logical value indicating whether the input is a single data frame with paired signatures or two data frames with unpaired signatures. The default is FALSE.
3. cell_lines: A character vector of cell lines to filter the consensus list to. The default is NULL, which means no filtering.
4. cutoff: The absolute cutoff value of similarity to use when filtering the consensus list. The default is 0.321.

# Get the consensus concordant signatures for the upregulated signature

consensus <- consensus_concordants(up_concordants, dn_concordants, paired = TRUE, cutoff = 0.2)

consensus |>
    head() |>
    knitr::kable()

Alternate One-Step Method

The above method breaks down the entire method into five steps. However, drugfindR also provides two functions that perform the entire pipeline in one function call with sensible defaults. These functions are investigate_target() and investigate_signature().

For this use case, investigate_signature() is the function that we want to use. It takes the following required arguments:

1. expr: The signature to investigate.
2. output_lib: The iLINCS library to target for concordant signatures. This can be one of c(“OE”, “KD”, “CP”), standing for overexpression, knockdown, and chemical perturbagens, respectively.
3. filter_threshold: The absolute threshold (or a pair of absolute thresholds) for the logFC values. Any genes that do not meet the threshold will be removed from the signature.
4. filter_prop: The percentile threshold (or a pair of percentile thresholds) for the logFC values. Any genes that do not meet the threshold will be removed from the signature.

Other arguments that have sensible defaults are:

1. similarity_threshold: The absolute cutoff value of similarity to use when filtering the consensus list. The default is 0.2.
2. paired: A logical value indicating whether the to split the input dataframe in up and downregulated signatures. The default is TRUE.
3. output_cell_lines: A character vector of cell lines to filter the consensus list to. The default is NULL, which means no filtering.
4. gene_column: The name of the column in the input that contains the gene names. The default is "Symbol".
5. logfc_column: The name of the column in the input that contains the log fold change values. The default is "logFC".
6. pval_column: The name of the column in the input that contains the p-values. The default is "PValue".
7. source_name: The name of the source of the signature. The default is "Input".
8. source_cell_line: The cell line of the source of the signature. The default is "NA".
9. source_time: The time of the source of the signature. The default is "NA".
10. source_concentration: The concentration of the source of the signature. The default is "NA".

investigated <- investigate_signature(diffexp,
    output_lib = "CP", filter_threshold = 1.5,
    gene_column = "hgnc_symbol", logfc_column = "logFC", pval_column = "PValue"
)

investigated |>
    head() |>
    knitr::kable()