Sets the normalization of a ScreenR object, so that samples sequenced at different depths can be compared. The raw counts are not changed: the normalization is the effective library size of each sample, and the normalized count of a barcode is its raw count divided by the effective library size of its sample, times 1e6 (counts per million). Three methods are available:
"cpm" (default): the effective library size is the total number of reads of the sample;
"tmm": the total number of reads multiplied by the trimmed mean of M-values (TMM) normalization factor of edgeR (
normLibSizes). This is more robust when strong hits deplete a large part of the library;"control": the number of reads of the control barcodes (for example non-targeting shRNAs, or the sgRNAs of non-essential genes) of the sample. This assumes that the control barcodes are not affected by the treatment.
Usage
normalize_data(
screenR_Object,
method = c("cpm", "tmm", "control"),
control_barcodes = NULL
)Arguments
- screenR_Object
The ScreenR object obtained using
create_screenr_object.- method
The normalization method: "cpm" (default), "tmm" or "control".
- control_barcodes
The identifiers of the control barcodes for the "control" method (required by that method, ignored by the others). Identifiers that are not in the object give a warning; if none is in the object, it is an error.
Details
The normalization is used consistently by the whole analysis:
the normalized counts (get_normalized_count_table), the
data table (get_data_table) and everything computed from
it (compute_metrics, plot_trend,
plot_boxplot, compute_slope, ...), the
sample plots (plot_mds, plot_pca,
plot_replicate_correlation) and the edgeR and limma
models (create_edger_obj, used by
find_camera_hit, find_roast_hit,
compute_barcode_statistics, find_time_hit,
filter_by_variance and plot_barcode_hit),
whose offsets are the logarithms of the effective library sizes. The
models always use the raw counts. Logarithms and pseudocounts used by
the plots (for example log2(normalized + 1) in plot_pca)
are display transformations: they do not change the normalization.
It is stored in the object: the effective library sizes in the
EffectiveLibSize column of the colData, the method and its provenance
in the metadata (see get_normalization). Replacing the
counts removes it; subsetting the object keeps the effective library
sizes of the retained samples (call normalize_data() again to
normalize on the retained barcodes only). Calling normalize_data()
again replaces the normalization, and every view of the object follows
immediately.
Every sample must have reads (and, for "control", reads of the control barcodes): a sample with an effective library size of zero cannot be normalized and is an error.
Choosing the method. In a dropout screen many barcodes are depleted, so with "cpm" all the other barcodes look enriched. "tmm" reduces this bias, but it assumes that most barcodes do not change and that the changes go in both directions: when many barcodes change in the same direction it corrects only part of the bias (in a simulation with 20% of the genes depleted 8 times it removed about half of it). "control" removes it when control barcodes known to be without effect are available (for example the sgRNAs of non-essential genes).
Examples
object <- get0("object", envir = asNamespace("ScreenR"))
object <- normalize_data(object)
get_normalization(object)$method
#> [1] "cpm"
get_normalized_count_table(object)
#> ScreenR normalized count table: 5320 rows, 15 columns
#> # A tibble: 5,320 × 15
#> Barcode T1 T2 Time3_TRT_A Time3_TRT_B Time3_TRT_C Time3_A Time3_B
#> <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
#> 1 10208 330. 437. 365. 439. 311. 358. 425.
#> 2 10502 26.2 34.6 20.5 59.5 16.7 49.3 17.4
#> 3 10796 141. 156. 164. 124. 132. 120. 121.
#> 4 11089 356. 424. 336. 424. 375. 388. 346.
#> 5 11382 67.2 65.7 82.6 36.9 26.3 58.4 65.4
#> 6 11675 687. 775. 722. 682. 669. 820. 797.
#> 7 11969 455. 548. 623. 540. 558. 488. 438.
#> 8 12263 144. 119. 112. 138. 95.0 158. 128.
#> 9 12555 406. 413. 459. 518. 483. 408. 425.
#> 10 12847 43.7 42.6 30.2 34.9 29.6 34.6 20.5
#> # ℹ 5,310 more rows
#> # ℹ 7 more variables: Time3_C <dbl>, Time4_TRT_A <dbl>, Time4_TRT_B <dbl>,
#> # Time4_TRT_C <dbl>, Time4_A <dbl>, Time4_B <dbl>, Time4_c <dbl>
# TMM normalization: the same effective library sizes are used by the
# plots and by the edgeR models
object <- normalize_data(object, method = "tmm")
get_normalization(object)$effective_library_size
#> T1 T2 Time3_TRT_A Time3_TRT_B Time3_TRT_C Time3_A
#> 10853404 9827013 6840595 9229604 9737088 9647737
#> Time3_B Time3_C Time4_TRT_A Time4_TRT_B Time4_TRT_C Time4_A
#> 9179789 9641856 8799318 9717098 12044656 10302865
#> Time4_B Time4_c
#> 9128760 9182589
create_edger_obj(object)$samples
#> group lib.size norm.factors
#> T1 T1/T2 10853404 1
#> T2 T1/T2 9827013 1
#> Time3_TRT_A Treated 6840595 1
#> Time3_TRT_B Treated 9229604 1
#> Time3_TRT_C Treated 9737088 1
#> Time3_A Control 9647737 1
#> Time3_B Control 9179789 1
#> Time3_C Control 9641856 1
#> Time4_TRT_A Treated 8799318 1
#> Time4_TRT_B Treated 9717098 1
#> Time4_TRT_C Treated 12044656 1
#> Time4_A Control 10302865 1
#> Time4_B Control 9128760 1
#> Time4_c Control 9182589 1