def triu_plot(x, y, color, label, **kwargs):
z = QCplot.density_interpolate(x, y)
idx = z.argsort()
x, y, z = x[idx], y[idx], z[idx]
plt.scatter(x, y, c=z, **kwargs)
plt.axhline(0, ls=":", lw=0.1, c="#484848", zorder=0)
plt.axvline(0, ls=":", lw=0.1, c="#484848", zorder=0)
(x0, x1), (y0, y1) = plt.xlim(), plt.ylim()
lims = [max(x0, y0), min(x1, y1)]
plt.plot(lims, lims, ls=":", lw=0.1, c="#484848", zorder=0)
def downsample_sgrnas(
counts,
lib,
manifest,
n_guides_thresholds,
n_iters=10,
plasmids=None,
gene_col="Gene",
):
plasmids = ["CRISPR_C6596666.sample"] if plasmids is None else plasmids
# Guides library
glib = (
lib[lib.index.isin(counts.index)].reset_index()[["sgRNA_ID", gene_col]].dropna()
)
glib = glib.groupby(gene_col)
scores = []
for n_guides in n_guides_thresholds:
for iteration in range(n_iters):
LOG.info(f"Number of sgRNAs: {n_guides}; Iteration: {iteration + 1}")
# Downsample randomly guides per genes
sgrnas = pd.concat(
[d.sample(n=n_guides) if d.shape[0] > n_guides else d for _, d in glib]
).set_index("sgRNA_ID")
# sgRNAs fold-change
sgrnas_fc = counts.loc[sgrnas.index].norm_rpm().foldchange(plasmids)
# genes fold-change
genes_fc = sgrnas_fc.groupby(sgrnas[gene_col]).mean()
genes_fc = genes_fc.groupby(manifest["model_id"], axis=1).mean()
# AROC
res = pd.DataFrame(
[
dict(sample=s, aroc=QCplot.aroc_threshold(genes_fc[s], fpr_thres=.2)[0])
for s in genes_fc
]
).assign(n_guides=n_guides)
scores.append(res)
return pd.concat(scores)
def guides_recall_benchmark(metrics,
sgrna_counts,
dataset,
smap,
nguides_thres=None,
jacks_thres=1.,
fpr_thres=0.01):
nguides_thres = [1, 2, 3, 4, 5, 100
] if nguides_thres is None else nguides_thres
# Define set of guides
LOG.info(f"#(sgRNAs)={metrics.shape[0]}")
# AROC scores
scores = []
for m in ["KS", "JACKS_min", "combined"]:
LOG.info(f"Metric = {m}")
for n in nguides_thres:
# Metric top guides
if m == "combined":
metric_topn = (
metrics.query(f"JACKS_min < {jacks_thres}").sort_values(
"KS",
ascending=False).groupby("Approved_Symbol").head(n=n))
else:
metric_topn = (metrics.sort_values(
m, ascending=(
m == "JACKS_min")).groupby("Approved_Symbol").head(
n=n))
# Calculate fold-changes on subset
metric_fc = sgrna_counts.loc[metric_topn.index]
metric_fc = metric_fc.norm_rpm().foldchange(dataset.plasmids)
metric_fc = metric_fc.groupby(metrics["Approved_Symbol"]).mean()
metric_fc = metric_fc.groupby(smap["model_id"], axis=1).mean()
# Binarise fold-changes
metric_thres = [
QCplot.aroc_threshold(metric_fc[s], fpr_thres=fpr_thres)[1]
for s in metric_fc
]
metric_bin = (metric_fc < metric_thres).astype(int)
genes, samples = set(metric_bin.index), set(metric_bin)
LOG.info(f"Genes:{len(genes)}; Samples:{len(samples)}")
# Evaluation
metric_recalls = pd.DataFrame([
dict(
sample=s,
metric=m,
nguides=n,
ess_aroc=QCplot.aroc_threshold(metric_fc.loc[genes, s],
fpr_thres=.2)[0],
recall=recall_score(ky_bin.loc[genes, s],
metric_bin.loc[genes, s]),
precision=precision_score(ky_bin.loc[genes, s],
metric_bin.loc[genes, s]),
) for s in metric_bin
])
# Store
scores.append(metric_recalls)
scores = pd.concat(scores)
return scores
master_lib["JACKS_min"] = abs(master_lib["JACKS"] - 1)
master_lib = master_lib[master_lib.index.isin(ky_counts.index)]
# Project Score KY v1.1: Fold-changes
#
FDR_THRES = 0.01
ky_sgrna_fc = ky_counts.loc[master_lib.index].norm_rpm().foldchange(
ky.plasmids)
ky_fc = (ky_sgrna_fc.groupby(master_lib["Approved_Symbol"]).mean().groupby(
ky_smap["model_id"], axis=1).mean())
ky_thres = [
QCplot.aroc_threshold(ky_fc[s], fpr_thres=FDR_THRES)[1] for s in ky_fc
]
ky_bin = (ky_fc < ky_thres).astype(int)
# Benchmark sgRNA: Essential/Non-essential AROC
#
metrics_recall = guides_recall_benchmark(master_lib,
ky_counts,
ky,
ky_smap,
fpr_thres=FDR_THRES,
jacks_thres=1.)
metrics_recall.to_excel(f"{RPATH}/KosukeYusa_v1.1_benchmark_recall.xlsx",
index=False)
data_export = ky.counts.copy()
data_export.insert(0, "MinLibCas9_guide", data_export.index.isin(minlibcas9.index))
data_export.insert(
0, "Approved_Symbol", kylib.reindex(data_export.index)["Approved_Symbol"]
)
data_export.to_excel(f"{RPATH}/GuideCoverage_export_data.xlsx")
# Essential genes AROC
#
for l in libraries:
libraries[l]["aurc"] = pd.Series(
{
c: QCplot.aroc_threshold(libraries[l]["fc"][c], fpr_thres=0.2)[0]
for c in libraries[l]["fc"]
}
)
# Replicates correlation
#
for l in libraries:
libraries[l]["reps"] = replicates_correlation(
libraries[l]["fc_rep"].rename(columns=ky_ss["name"]), method="spearman"
)
# Essential genes AURC
]:
plot_df = pd.concat(
[
mlib[["Library", mtype]], minlib[[mtype
]].assign(Library="MinLibCas9")
],
ignore_index=True,
).dropna()
fig, ax = plt.subplots(1, 1, figsize=(1.5, 2), dpi=600)
QCplot.bias_boxplot(
plot_df,
x="Library",
y=mtype,
add_n=False,
tick_base=None,
order=order,
draw_violin=True,
ax=ax,
)
ax.set_xticklabels(order, rotation=45, ha="right")
ax.grid(True, ls=":", lw=0.1, alpha=1.0, zorder=0, axis="y")
ax.set_xlabel("")
ax.set_ylabel(f"{mtype.split('_')[1]}")
plt.savefig(
f"{RPATH}/lib_metrics_library_boxplot_{mtype}.pdf",
bbox_inches="tight",
# sgRNA fold-changes
fc = ky_counts.copy().loc[lib.index].norm_rpm().foldchange(ky.plasmids)
# Gene fold-changes
fc_gene = fc.groupby(lib["Approved_Symbol"]).mean()
# Gene average fold-changes
fc_gene_avg = fc_gene.groupby(ky_smap["model_id"], axis=1).mean()
# Gene average scaled fold-changes
fc_gene_avg_scl = ReadCounts(fc_gene_avg).scale()
# FDR threshold
fpr = pd.DataFrame(
{
s: QCplot.aroc_threshold(fc_gene_avg[s], fpr_thres=fdr_thres)
for s in fc_gene_avg
},
index=["auc", "thres"],
).T
fpr["auc"] = [
QCplot.aroc_threshold(fc_gene_avg[s], fpr_thres=0.2)[0]
for s in fpr.index
]
# Gene binarised
fc_gene_avg_bin = (fc_gene_avg[fpr.index] < fpr["thres"]).astype(int)
# Store
libraries[ltype]["fc"] = fc
libraries[ltype]["fc_gene"] = fc_gene
)
for i, mtype in enumerate(row_order):
plot_df = pd.concat(
[polyt_pos_df, master_lib.loc[polyt_pos_df.index, mtype]], axis=1)
for j, p in enumerate(polyt):
ax = axs[i][j]
order = natsorted(set(plot_df[p].dropna().astype(int)))
QCplot.bias_boxplot(
plot_df[plot_df["polyt5"].isnull()]
if p == "polyt4" else plot_df[~plot_df["polyt5"].isnull()],
x="polyt4",
y=mtype,
add_n=True,
tick_base=None,
order=order,
ax=ax,
)
ax.set_xticklabels(ax.get_xticklabels() if i == (len(row_order) -
1) else [],
rotation=0,
horizontalalignment="right")
ax.grid(True, ls=":", lw=0.1, alpha=1.0, zorder=0, axis="y")
if mtype == "JACKS":
ax.axhline(1,
ls="-",
plt.close("all")
# Recall gene lists
#
gsets_aucs = {}
for n, gset in [
("essential", Utils.get_essential_genes()),
("non-essential", Utils.get_non_essential_genes()),
]:
# Aroc
plt.figure(figsize=(2, 2), dpi=600)
ax = plt.gca()
_, stats_ess = QCplot.plot_cumsum_auc(fc_gene[samples],
gset,
palette=sample_pal,
legend_prop={"size": 4},
ax=ax)
plt.title(f"{n} recall curve")
plt.xlabel("Percent-rank of genes")
plt.ylabel("Cumulative fraction")
plt.grid(True, ls=":", lw=0.1, alpha=1.0, zorder=0, axis="both")
plt.savefig(f"{RPATH}/minlibcas9_screens_roccurves_{n}.pdf",
bbox_inches="tight")
plt.close("all")
# Barplot
df = pd.Series(stats_ess["auc"])[samples].rename("auc").reset_index()
plt.figure(figsize=(3, 0.15 * len(samples)), dpi=600)
sns.barplot(
# sgRNA fold-changes
fc = ky_counts.copy().loc[lib.index].norm_rpm().foldchange(ky.plasmids)
# Gene fold-changes
fc_gene = fc.groupby(lib["Approved_Symbol"]).mean()
# Gene average fold-changes
fc_gene_avg = fc_gene.groupby(ky_smap["model_id"], axis=1).mean()
# Gene average scaled fold-changes
fc_gene_avg_scl = ReadCounts(fc_gene_avg).scale()
# FDR threshold
fpr = pd.DataFrame(
{
s: QCplot.aroc_threshold(fc_gene_avg[s], fpr_thres=FDR_THRES)
for s in fc_gene_avg
},
index=["auc", "thres"],
).T
fpr["auc"] = [
QCplot.aroc_threshold(fc_gene_avg[s], fpr_thres=0.2)[0] for s in fpr.index
]
# Gene binarised
fc_gene_avg_bin = (fc_gene_avg[fpr.index] < fpr["thres"]).astype(int)
# Store
libraries[ltype]["fc"] = fc
libraries[ltype]["fc_gene"] = fc_gene
libraries[ltype]["fc_gene_avg"] = fc_gene_avg
libraries["Minimal"]["fc_gene_avg"].index)
LOG.info(f"Genes={len(genes)}")
# Essential/non-essential AROC and AURC
#
metrics_arocs = []
for ltype in libraries:
for s in libraries[ltype]["fc_gene_avg"]:
LOG.info(f"Library={ltype}; Organoid={s}")
metrics_arocs.append(
dict(
library=ltype,
sample=s,
aroc=QCplot.aroc_threshold(libraries[ltype]["fc_gene_avg"][s],
fpr_thres=0.2)[0],
))
metrics_arocs = pd.DataFrame(metrics_arocs)
# Plot essential genes recall
#
pal = dict(All="#e34a33", Minimal="#fee8c8")
n = libraries["All"]["fc_gene_avg"].shape[1]
fig, ax = plt.subplots(1, 1, figsize=(0.6 * n, 2.0), dpi=600)
sns.barplot("sample",
"aroc",
"library",
ky_counts = ky.counts.remove_low_counts(ky.plasmids)
ky_fc = ky_counts.norm_rpm().norm_rpm().foldchange(ky.plasmids)
ky_gsets = define_sgrnas_sets(ky.lib, ky_fc, add_controls=True)
master_lib = Library.load_library("MasterLib_v1.csv.gz").query(
"Library == 'KosukeYusa'")
# sgRNAs sets AURC
#
for m in ky_gsets:
LOG.info(f"AURC: {m}")
ky_gsets[m]["aurc"] = pd.Series({
s: QCplot.recall_curve(ky_fc[s], index_set=ky_gsets[m]["sgrnas"])[2]
for s in ky_fc
})
ky_gsets_aurc = pd.concat([
ky_gsets[m]["aurc"].rename("aurc").to_frame().assign(dtype=m)
for m in ky_gsets
])
ky_gsets_aurc.to_excel(f"{RPATH}/ky_v11_guides_aurcs.xlsx")
# sgRNA sets histograms
#
plt.figure(figsize=(2.5, 1.5))
for c in ky_gsets:
sns.distplot(