Skip to contents

Changelog

Source: NEWS.md

ScreenR 1.15.16

  • New plot_volcano(): every gene with the mean log2 fold change of its barcodes and its -log10 FDR, from the table of one method or of summarise_hits() (with the barcode statistics); the most significant hits are labelled and genes circles the genes to follow. Used in Tutorials 3 and 4.

  • New ScreenR app, run_screenr_app(): the analysis of a screen in the web browser without writing code, for the people who did the experiment. It loads an example screen, a count table or a MAGeCK count file, describes the samples, shows the quality control, finds the hits with FRY, CAMERA and the Z-score tests, and shows each hit at the level of its barcodes; tables and an R script that repeats the analysis can be downloaded. It needs shiny, bslib and DT (suggested packages). See Tutorial 5.

  • New design of the pkgdown site: frosted glass panels and a floating navbar over a soft colour field, with a light that follows the cursor and a light/dark/auto switch. The home page has highlights and buttons, the tutorials a progress indicator with previous/next links, the articles a reading progress bar; figures can be enlarged, Cmd/Ctrl + K opens the search, the reference index is grouped in cards, and a pair of scissors cuts the DNA of the logo. Animations stop with the reduced motion setting of the system, and the site has a card for links shared on social media.

  • The README, the pkgdown home page, the DESCRIPTION and the package help describe shRNA and CRISPR-Cas9 screens. The README quick start uses a sample table, palettes, theme_set(), FRY, CAMERA and the empirical null, and adds a CRISPR dropout example; its method table follows Tutorial 3. The workflow diagram shows the CRISPR steps and the tests across time points.

  • plot_mds() and plot_pca() leave room for the labels of the samples at the edges of the plot.

  • plot_treatment_separation() plots a time course: when condition_column and time_column are the same variable (for example “Day” in a dropout screen), treatment can be several time points and each is compared with the control time point, with every time point on the x axis. Previously this found no shared stage and gave an error. With real time the labels of the first and last time points are no longer clipped.

  • compute_metrics(condition_column = "Day") compares two time points (control = "T0", treatment = "T18"), for dropout screens without a treatment; day is then not used. The default compares a treatment and a control at the time point day, as before.

  • Dropout screens (Tutorial 4, the empirical null and the MAGeCK benchmark vignettes, the examples) are described with the time points in the Daya column and a single Treatment, "Untreated", for all the samples.

  • Tutorial 4 checks the lost sgRNAs (barcode_lost(), plot_barcode_lost(), plot_barcode_lost_for_gene()), compares the time points with plot_zscore_distribution(), compute_slope() and plot_treatment_separation(), uses find_common_hit() and find_consistent_hit(), and inspects the sgRNAs of ten hits with plot_barcode_hit(), plot_barcode_trend(), plot_barcode_heatmap(), find_discordant_barcodes() and filter_by_variance().

  • Tutorial 4 estimates the efficiency of the sgRNAs from the depletion of the core essential genes, checks that it is reproducible between independent cultures and relates it to the GC content and the nucleotide next to the PAM.

  • Tutorial 3 (finding and validating hits) uses the current functions: a sample table and palettes at creation, theme_set(), TMM normalization, FRY as the recommended test, the empirical null, plot_common_hit() with any number of methods (UpSet plot), find_consistent_hit(), find_time_hit() with real times, and plot_trend(color_by = ), plot_treatment_separation() and plot_barcode_heatmap() to check the candidates.

  • Tutorial 4 (CRISPR-Cas9 screens) uses palettes, theme_set(), the trend of the reference genes over the real days, their sgRNAs with plot_barcode_heatmap(), the overlap of the methods with plot_common_hit(), and explains how the baseline of the normalization affects the self-contained tests.

  • plot_treatment_separation() plots descriptive treatment/control abundance ratios for any two condition labels and configurable time metadata. It retains all selected stages, supports real times and sample subsets, and inherits the user theme. Missing comparisons are never assigned zero effects.

  • plot_trend(color_by = "Treatment") overlays separate condition trends with biological sample points and lines joining time-point means. The original ungrouped regression remains the default without color_by; trend = "lm" explicitly fits separate regressions. Object palettes are used, and reference_groups shows shared baselines once as neutral points. connect_reference joins the shared reference history to condition means; highlight_difference shades the observed gap between two conditions. No theme is applied by these new plots; use theme_set() in user code.

  • plot_barcode_heatmap() shows individual barcode abundances across samples, with gene and sample-variable panels. Display log2 normalized counts, barcode-centered values or values relative to explicit reference samples.

  • Model designs now carry sample identifiers and are checked against the object before fitting. Custom designs must have sample identifiers as row names in object order. Named contrast vectors and one-column contrast matrices are aligned to coefficient names; missing, duplicated and unknown names are rejected. Unnamed numeric contrast vectors remain positional.

  • Package checks, coverage and website builds now run on pushes and pull requests to devel. Website builds on devel do not deploy GitHub Pages.

  • find_consistent_hit() tests directional effects shared by all supplied contrasts using FRY, ROAST, CAMERA or gene-level Z-scores (random or empirical reference). combine_consistent_hits() accepts complete one-sided gene-level p-value tables, combines their raw p-values with an intersection-union test and applies BH or BY correction across genes. Missing or duplicated genes and invalid p-values are errors.

  • find_camera_hit() and find_roast_hit() gain keep_all = TRUE to retain opposite observed directions with their requested one-sided p-values; defaults are unchanged. The longitudinal vignette uses the built-in consistency test.

  • Palettes: create_screenr_object(palettes = ) and the new set_palettes()/get_palettes() store colors for the values of the sample variables (Treatment, Group, Day, Sample, …), checked against the object and kept when it is subset. Every plot that colors the samples by one of these variables uses them: plot_mapped_reads(), plot_mapped_reads_distribution(), plot_barcode_lost(), plot_barcode_lost_for_gene(), plot_gini(), plot_mds(), plot_pca() and plot_boxplot(). An explicit palette argument replaces the palette of the object. plot_mds() and plot_pca() accept the name of a sample variable in groups (for example groups = "Treatment"). plot_zscore_distribution(), which receives the tables of compute_metrics() rather than the object, has a new palette argument (for example palette = get_palettes(object)$Day).

  • The longitudinal vignette sets palettes at the creation of the object and uses TMM normalization throughout.

  • plot_common_hit() has been rewritten: it takes the results of any number of methods, from any hit-finding function of ScreenR (find_gene_zscore_hit(), find_camera_hit(), find_roast_hit(), find_time_hit(), find_zscore_hit(), …) or vectors of gene names, as separate arguments or as one named list, with an optional fdr threshold for results computed with fdr = 1. Up to three methods give a Venn diagram, more an UpSet plot (type chooses). Both are drawn with ggplot2 and patchwork and apply no theme, so ScreenR no longer depends on ggvenn. Barcode-level tables are refused. Calls of previous versions keep working (unnamed results are now labelled with the expression given, except hit_zscore, hit_camera and hit_roast, which keep their labels).

  • find_common_hit() takes the same input as plot_common_hit(): any number of methods, with common_in (all the methods by default) and fdr.

  • The longitudinal vignette compares five methods at each time point with UpSet plots, ranks the candidates with explicit rules (agreement of the self-contained tests, number of shRNAs, discordant shRNAs, dependence on one replicate, fitness, time trend) and ends with the hits to validate and their plots at T3 and T4. It sets the ggplot2 theme with theme_set().

  • find_time_hit() has a new argument time, the real time of each time point (for example time = c(Time3 = 7, Time4 = 14)), validated as in compute_slope() and plot_trend(). The model is then fitted against the actual time and Log2FC is the change of the treatment effect per unit of time. By default (time = NULL) the time points are numbered 0, 1, 2, … as before, so existing results do not change; with two time points the p-values are the same with or without real times.

  • compute_slope(), filter_by_slope() and plot_trend() reject times that are not finite.

  • New vignette “Analysis of a longitudinal screen: from counts to treatment-specific hits”, a complete analysis of the example screen (T0, post-puromycin baseline, control and treated cultures at two time points with three replicates): quality control, one normalization for the whole analysis, one edgeR model with all the samples, treatment effects at each time point, the difference between depletion at both time points (intersection-union test), on average and changing over time (find_time_hit()), fitness genes shared by all the cultures, evidence at the level of the shRNAs, a sensitivity analysis and the current limitations of ScreenR for longitudinal designs.

One source of truth for the data

  • The ScreenR object stores each piece of information once, in its SummarizedExperiment parts: the raw counts in assay(object, "counts") (one row per barcode, one column per sample, the barcode identifiers as row names), the annotation of the barcodes in the rowData, the sample information, groups and per-sample replicate labels in the colData. The normalized counts and the data table are no longer stored: they are computed from these parts when needed, so they cannot get out of date. compute_data_table() only checks that the object is normalized and is no longer necessary.
  • Barcode identifiers are always character and never become count columns: numeric identifiers are converted to their decimal form (10208 and “10208” are the same barcode, 1e5 is “100000”), in the count table, the annotation and the control barcodes.
  • Stricter input checks in create_screenr_object() and set_count_table(): missing identifiers, and missing, infinite or negative counts are errors; duplicated barcodes are an error unless duplicate_barcodes = "sum" or "first" (the example count_table has three “*” rows: remove them, as the tutorials do); barcodes annotated more than once with different values are an error unless duplicate_annotation = "first" (exact copies are merged).
  • The object is checked (validity method): unique identifiers, a Barcode column of the rowData equal to the row names, valid counts and a complete normalization. assay<-, assays<-, rowData<-, colData<- (and $<-) and dimnames<- keep these rules (see ?"screenr_object-replace").
  • Changes propagate: replacing the counts with different values (with set_count_table() or assay<-) removes the normalization; normalizing again changes every view at once; replacing the annotation (set_annotation_table() or rowData<-) changes the gene of each barcode in all the following analyses; subsetting and reordering keep the barcodes, the annotation and the samples aligned by identifier.
  • get_annotation_table() returns the annotation of the barcodes of the object (rowData) followed by the annotation rows of barcodes that are not in the count table, which are kept apart. barcodes_per_gene() counts the barcodes of the object.
  • set_data_table() is deprecated: the data table cannot be replaced (it only accepts the current data table, with a warning). set_normalized_count_table() only accepts tables that scale each sample by one factor, stored as effective library sizes (method “custom”).

One normalization for the whole analysis

  • normalize_data() no longer stores normalized counts: it records the effective library size of each sample in colData(object)$EffectiveLibSize and the method, the control barcodes and the provenance in metadata(object)$normalization (see the new get_normalization()). The raw counts are never changed.
  • The same normalization is used everywhere: the normalized counts, the data table and everything computed from it (compute_metrics(), the plots, compute_slope(), …), plot_mds(), plot_pca(), plot_explained_variance(), plot_replicate_correlation() and the edgeR and limma models. create_edger_obj(), used by find_camera_hit(), find_roast_hit() (ROAST and FRY), compute_barcode_statistics(), find_time_hit(), filter_by_variance() and plot_barcode_hit(), builds the DGEList from the raw counts with the effective library sizes as library sizes (normalization factors 1), so the offsets of the models are the same normalization as the plots and TMM is applied once.
  • Before, the edgeR/limma based methods always used the total counts of the object (CPM), whatever the method of normalize_data(), and plot_replicate_correlation() used CPM too. Results change with method = "tmm" or "control": in the HAP1 screen (day 18 vs T0) FRY calls 694 genes with TMM and 979 with the non-essential sgRNAs as controls at 5% FDR, where it called 721 with any method before. With the default "cpm" the results are unchanged, except when the object was subset after normalization: the models now use the library sizes of the normalization instead of the totals of the retained barcodes (small changes, at most 0.09 in log10 p-value in a test on 300 genes).
  • Operations that need normalized counts (data table, metrics, trend and sample plots, statistical methods) stop with an informative error when the object is not normalized, instead of using empty tables or normalizing silently (plot_mds(), plot_pca() and plot_explained_variance() used CPM silently). Raw-count quality control (compute_gini(), mapped_reads(), barcode_lost(), …) does not need normalization.
  • Zero library sizes, and control barcodes that are missing (all of them) or have no reads in a sample, are errors in normalize_data(); control barcodes partly missing give a warning.
  • ?normalize_data explains that TMM corrects only part of the compositional bias when many barcodes change in the same direction (about half in a simulation with 20% of the genes depleted 8 times), while control barcodes remove it.

Objects of older versions

  • update_screenr_object() (called by all the functions) converts objects created before 1.15.8 and objects of versions 1.15.8 to 1.15.15. Their normalized counts and data table are not kept, because their normalization was not recorded: run normalize_data() on the converted object (a message says so). The internal example object has been converted and normalized with “cpm” (its normalized counts are identical to the old ones).

ScreenR 1.15.15

  • New tutorial “Tutorial 4: CRISPR-Cas9 screens”: installing CRISPRcleanR, describing a time-course dropout screen, choosing the sgRNA thresholds, checking the screen with reference genes, correcting copy-number bias with correct_crispr_bias(), finding the hits with the non-essential genes as reference, and positive-selection screens.
  • New datasets hap1_count_table and hap1_annotation_table: a subset of the HAP1 TKOv3 CRISPR-Cas9 knockout screen of Hart et al. (2017), with the genomic position of each sgRNA.
  • correct_crispr_bias() documents that CRISPRcleanR must be run on the complete genome-wide library: on a subset of genes enriched in hits it removes real signal.
  • Tutorial 4 adds FRY (find_roast_hit()) to the comparison of the methods (with a single T0 sample it works, as the variability is estimated from the replicates of the other time points), shows how conservative the methods without reference are on this subset enriched in essential genes, and explains the positive fold change of the non-essential genes after CPM normalization (normalizing to their sgRNAs puts them back at 0).
  • find_discordant_barcodes() documents that it is for inspecting the hits, not for filtering the barcodes before finding them: the filter is circular and, in the K562 screen, finds fewer essential and more non-essential genes.
  • Tutorial 4 shows the general quality control plots on the HAP1 screen (plot_mapped_reads(), plot_mapped_reads_distribution(), plot_mds(), plot_explained_variance() and plot_boxplot()).
  • plot_boxplot() orders the fill legend as the samples (T0, T3, T18), not alphabetically (T0, T18, T3).
  • plot_barcode_hit() is drawn with ggplot2 instead of limma::barcodeplot() and returns a ggplot object (it returned NULL), so it can be saved, themed and combined with other plots. No theme is applied. It has the same enrichment line and barcode strip, the barcodes are colored by their log2 fold change, and a new panel (show_log2fc, TRUE by default) shows the log2 fold change of all the barcodes with those of the gene.
  • plot_reference_separation() no longer applies theme_bw(): set the theme with ggplot2::theme_set() or by adding it to the plot.
  • Clearer axis labels: “Explained variance” and “Principal component” in plot_explained_variance() (it said “Expressed Variance”), the dimensions of plot_mds() with the share of variance they explain, “Mapped reads” in plot_mapped_reads(), “Reads per barcode” in plot_mapped_reads_distribution(), “Lost barcodes” in plot_barcode_lost() and plot_barcode_lost_for_gene(), “Normalized counts” in plot_boxplot() and “Mean normalized counts” in plot_trend() (they said “Frequency”).
  • plot_trend() and plot_barcode_trend() use linewidth instead of the size aesthetic deprecated in ggplot2 3.4.0, so they no longer give a deprecation warning.
  • New plot_pca(): the samples on two principal components (PC1 and PC2 by default, components selects others), with the share of the variance on the axes.
  • plot_explained_variance() computed the principal components on the raw counts, so the first component mostly followed the sequencing depth of the samples (correlation 0.995 in the HAP1 screen). The principal components, and the MDS of plot_mds(), are now computed on the log2 normalized counts, with the normalization chosen in normalize_data() (“cpm”, “tmm” or “control”; counts per million when the counts have not been normalized).
  • plot_barcode_lost() drew the bars twice, an opaque layer under the one with alpha, so alpha had no visible effect; the bars are drawn once.
  • plot_trend() draws the samples at their time point instead of one position per sample, so replicates share the same time and the line is the one of compute_slope() (used by filter_by_slope()); before, the line was fitted along the order of the samples. New arguments of plot_trend(), compute_slope() and filter_by_slope(): time_column (the column with the time points, “Day” by default) and time (the real times of the time points, for example c(T0 = 0, T3 = 3, T18 = 18); by default they are equally spaced, as before).
  • plot_trend() and plot_boxplot() failed with several genes unless nrow or ncol was given; by default ggplot2 now chooses the layout.
  • find_gene_zscore_hit(reference = "empirical"): the gene scores are compared with an empirical null, the distribution of the genes without effect estimated from the centre of the distribution of the gene statistics (empirical_fit = "central", default, or "one-sided"). New plot_empirical_null() shows the fit. The random barcode sets of the default reference are conservative in dropout screens (the hits widen the null) and ignore that the barcodes of a gene vary together, which gives too small p-values in shRNA screens. In the genome-wide HAP1 CRISPR screen the empirical null found 90% (central) and 94% (one-sided) of the core essential genes at 5% FDR, against 58% with the default reference; in the K562 shRNA screen it kept the share of non-essential genes with a p-value below 0.05 under 4%, where the random barcodes reached 13% (default, with genome-wide proportions of essential genes) and 20% (non-essential genes as control_genes). See benchmark/realdata/empirical_null.R.
  • New vignette “Choosing the reference of the gene-level test”, which suggests reference = "empirical" and shows why on the HAP1 and K562 screens. The default of find_gene_zscore_hit() is still reference = "random", so existing results do not change.

ScreenR 1.15.14

CRISPR screens

ScreenR works with CRISPR-Cas9 screens (tested on the HAP1 TKOv3 screen of Hart et al. 2017, see the benchmark folder of the repository). New functions make it easier:

Documentation

  • The vignette “Benchmark against MAGeCK and other tools” adds BAGEL2, gscreend and two plain edgeR/limma workflows to the comparison, a CRISPR-Cas9 screen (HAP1, TKOv3) and the quality check of the K562 screen with the reference genes. Its claims are toned down to match the results: ScreenR controls the FDR and ranks the hits as well as the best tools, but is more conservative at a fixed FDR.

Other changes

ScreenR 1.15.13

  • Enriched genes: every hit-finding method accepts direction = "Up". find_zscore_hit() and find_robust_zscore_hit() gain a direction argument (default "Down", same results as before); find_camera_hit() and find_roast_hit() now check direction and give an error for a value other than "Down" or "Up" (a typo such as "up" used to return no genes silently).
  • New tests check that every method finds enriched and depleted genes without mixing them.

ScreenR 1.15.12

Documentation

Other changes

ScreenR 1.15.11

Documentation

  • New vignette “Benchmark against MAGeCK”: compares the ScreenR methods with MAGeCK RRA and MLE on simulated screens and on the K562 screen. MAGeCK and simulation results are precomputed in inst/extdata/benchmark; the ScreenR methods are run in the vignette. The scripts are in the benchmark folder of the GitHub repository.
  • find_gene_zscore_hit(): the documentation no longer calls control genes the best reference. Non-targeting shRNAs are often not a good model of a gene without effect (in the K562 screen they made hundreds of non-essential genes look depleted); check them before using control_genes.
  • find_roast_hit(): the documentation explains that ROAST and FRY need replicates in both groups and find almost no hits with a single control sample.

ScreenR 1.15.10

Calibrated p-values

A benchmark on simulated screens showed that the p-values of some methods were too conservative (with no true hits, fewer than 5% of the genes had p < 0.05).

  • find_camera_hit(), find_roast_hit() and find_time_hit() now report one-sided p-values for the requested direction. camera and roast report two-sided p-values, which ScreenR used as they were: they were twice too large. The FDR is computed from the one-sided p-values over all the genes.
  • find_gene_zscore_hit() is now a permutation test: the mean (or median, statistic) Z-score of the barcodes of each gene is compared with random sets of barcodes of the same size. The previous binomial test on the number of barcodes below a threshold had only a few possible p-values and was very conservative. The arguments zscore_threshold and the column NBarcodesHit are replaced by statistic, n_permutations, seed and the column Score.
  • find_gene_zscore_hit(control_genes = ) draws the random barcodes from control genes (for example non-targeting shRNAs), the best reference since it cannot contain true hits.
  • find_roast_hit(method = "fry") uses limma::fry(), an exact and fast version of ROAST without the limit of 1 / (nrot + 1) on the p-values.
  • fdr_method = "storey" in find_camera_hit(), find_roast_hit(), find_time_hit() and find_gene_zscore_hit() estimates the proportion of genes without effect and is less conservative than Benjamini-Hochberg when many genes have an effect.
  • The description of ROAST is corrected: it is a self-contained test, not a competitive one.

ScreenR 1.15.9

New dataset

  • k562_count_table, k562_annotation_table and k562_reference_genes: a subset of a published genome-wide shRNA essentiality screen in K562 cells (Morgens et al. 2016, Nat Biotechnol) with the reference core essential and non-essential genes of Hart et al. Since the true hits (the essential genes) are known, the dataset can be used to evaluate methods. The script that builds it is in inst/scripts and the licenses of the original data in inst/LICENSES_data.txt.

ScreenR 1.15.8

The ScreenR object is now a SummarizedExperiment

  • The screenr_object class now extends SummarizedExperiment: the raw and normalized counts are the assays counts and normalized, the barcode annotation is in rowData() and Sample, Day, Treatment and Group are in colData(). The object can be used with the Bioconductor tools that accept a SummarizedExperiment and subset like one (object[barcodes, samples]), keeping counts, normalized counts and data table consistent.
  • All the ScreenR functions, including the get_ and set_ functions, return the same results as before: every exported function was run on the example data with the previous and the new class and gave identical outputs.
  • Objects saved with older versions are converted automatically; update_screenr_object() converts them explicitly.
  • Code that reads the old slots directly needs to use the get_ functions: object@count_table -> get_count_table(object), object@normalized_count_table -> get_normalized_count_table(object), object@groups -> get_groups(object). The slots data_table and annotation_table are unchanged.
  • New dependencies: SummarizedExperiment and S4Vectors.

ScreenR 1.15.7

Documentation

  • Three new step-by-step tutorials: “1. Getting your data into ScreenR”, “2. Quality control” and “3. Finding and validating hits”.

ScreenR 1.15.6

Bug fixes (found by an independent review of 1.15.4)

  • find_time_hit() now includes a main effect for the treatment (log(abundance) = a + b * time + d * treated + c * time * treated), so that c measures how much the treatment effect changes over time. Previously the model assumed that treated and control samples start from the same level, so a constant treatment effect was reported as a growing one. The control and baseline samples must now cover at least two time points.
  • The documentation of summarise_hits() and the vignette no longer claim that the list of genes with MaxFDR <= alpha keeps the FDR: MaxFDR is a per-gene level-alpha intersection-union test and a ranking score, but the combined list has no guaranteed false discovery rate.
  • The examples of plot_barcode_hit() and filter_by_variance() (and some tests) combined model.matrix(~ groups) with makeContrasts(Treated - Control), which subtracts the intercept; they now use create_design_matrix() and create_contrast().
  • Several examples called compute_metrics() with treatment and control swapped; they now use treatment = "TRT".
  • find_camera_hit() and find_time_hit() no longer fail when only one gene can be tested, and give a clear error when no gene has more than number_barcode barcodes (also find_roast_hit()).
  • find_roast_hit(seed = ) restores the random number state of the session.
  • compute_metrics() warns when the MAD of the log2 fold changes is 0; find_discordant_barcodes() reports the barcodes of genes with MAD 0 that differ from the median.
  • find_time_hit() returns Direction as a factor, like the other methods.

Changes not listed in 1.15.4

Vignette

ScreenR 1.15.5

More general input

  • create_screenr_object() accepts tables with any column names: barcode_column and gene_column give the names of the barcode and gene columns, which are renamed internally to Barcode and Gene. get_column_names() returns the original names. When a column is not found the error suggests the most similar column and the argument to use.
  • The Day and Treatment of each sample can be given with sample_table (and sample_columns for its column names) instead of relying on sample names of the form Day_Treatment_Replicate. get_sample_table() shows how the samples are interpreted; a message is given when sample names do not follow the convention.
  • The input is checked when the object is created: groups must have one value for each sample; sample columns stored as text are converted to numbers; other non-numeric columns of the count table are removed with a message; duplicated barcodes and barcodes missing from the annotation give a warning; duplicated barcodes in the annotation are reduced to their first row.
  • The annotation columns Sequence, Library and Gene_ID are now optional.
  • New show() method, so that objects saved with older versions of ScreenR can still be printed.

New features

ScreenR 1.15.4

Bug fixes in the vignette

  • The design matrix of the example analysis was built with model.matrix(~ groups) and its columns renamed with unique(groups), which does not match the column order: the contrast “Time3_TRT” actually compared the Time3 control samples with the baseline. The vignette now uses create_design_matrix() and create_contrast() to compare treated and control samples of the same time point.
  • The “at least 2 methods” common hits used roast_hit_TRT$Day3 and $Day6, which do not exist, so ROAST was ignored.

Changes that affect the results

  • find_camera_hit() gains fdr = 0.05 and returns only the genes with FDR <= fdr (previously all the genes with the requested direction). Use fdr = 1 for the previous behaviour.
  • find_roast_hit() gains fdr = 0.05 and filters on the FDR; the raw p-value cut off p_val is now optional (default NULL, previously 0.05).
  • compute_metrics() gains pseudocount = 0.5: the log2 fold change is log2((Treated + pseudocount) / (Control + pseudocount)) instead of log2(Treated / Control + 1e-7), and only barcodes with zero reads in both treated and control are removed (previously all barcodes with zero control reads).

New features

ScreenR 1.15.3

Changes

  • find_zscore_hit() now checks the metric argument ("median", the default, or "mean") and gives a clear error for any other value.
  • find_robust_zscore_hit() now has a default number_barcode = 6, the same as find_zscore_hit(), and returns an ungrouped tibble like the other hit functions.
  • The thresh and lfc arguments of find_camera_hit() are deprecated. They never had an effect on the result; passing them now gives a warning. Filter the returned table on its FDR column instead.
  • Removed the stray README.knit.md file from the package.
  • Added tests for these changes and for the 1.15.2 bug fixes.

ScreenR 1.15.2

Bug fixes (these change the results of some analyses)

ScreenR 1.15.1

  • Fixed the vignette build failure with recent edgeR versions (DPOTRF error code -4 in find_camera_hit()).

ScreenR 0.99.53

  • Changes in version 0.99.53:

  • Update to the documentation for the example dataset as requested by the reviewer

  • Pushed to Bioconductor

  • Changes in version 0.99.52:

  • Pushed to Bioconductor

  • Changes in version 0.99.51:

  • Pushed to Bioconductor

  • Changes in version 0.99.5:

    • Now ScreenR uses R 4.2 or above
    • Lazy data is set to FALSE
    • In the DESCRIPTION is now present a BugReport link
    • Now all the function are imported using importFrom
    • The vignette has now the BiocStyle
    • All chunks of code of the vignette are set to show warnings
    • Implemented the set and get function for the ScreenR object

ScreenR 0.99.41

  • Changes in version 0.99.41:
    • Update to Bioconductor

ScreenR 0.99.4

  • Changes in version 0.99.4:
    • Changed dataframe for the example

ScreenR 0.99.3

  • Changes in version 0.99.3:
    • Fixed some other bug
    • Changed find_common_hit

ScreenR 0.99.2

  • Changes in version 0.99.2:
    • Fixed some other bug

ScreenR 0.99.1

  • Changes in version 0.99.1:
    • Fixed some bug

ScreenR 0.99.0

  • Changes in version 0.99.0:
    • Submitted to Bioconductor