Skip to content

This tutorial is still under construction

Tutorial 3: Imputation and Normalization

Imputation

Missing values are common in proteomics. scpviz provides several imputation methods.

Note

Pre-processing functions like normalize() and impute() act directly on the pAnnData object rather than returning a copy. By default, the processed data are written to the active .X layer unless a new layer name is specified.

KNN Imputation

here: include a figure showing what KNN imputation looks like

pdata.impute(method="knn", n_neighbors=5)

Group-wise Imputation

scpviz also supports group-wise imputation. we implement it like this (maybe show diagram?) e.g. for median method, with groupby condition: can just show the example we have in dev actually!

pdata.impute(method="group_mean", groupby="condition")

Checking Imputation Stats

pdata.stats["imputation"]

📊 This dictionary stores how many values were imputed per group or overall.

Normalization

Brief explanation of normalization - to try to remove the effects of small variation in sample prep/sample loading, or mass spec performance. The assumption is that all samples (if global) or all samples within the group (if use groupby) should have a similar distribution of abundances.

scpviz implements several options: (put as table?)

method how it works recommended use mean makes mean of all samples match median makes median of all samples match typically use in bulk proteomics sum makes sum of all samples match typically use in single cell rnaseq, but not as appropriate for scp? directlfq (look at paper) used for single cell proteomics (see paper link)

Normalization

# Normalize intensities by global median
pdata.normalize(method="median")

# Normalize to reference proteins
pdata.normalize(method="reference_feature", reference_columns=["ACTB", "GAPDH"])

💡 Note: Normalization choices can affect downstream DE and enrichment.