| Title: | Predicting Interactomes from Co-Elution |
|---|---|
| Description: | PrInCE (Predicting Interactomes from Co-Elution) uses a naive Bayes classifier trained on dataset-derived features to recover protein-protein interactions from co-elution chromatogram profiles. This package contains the R implementation of PrInCE. |
| Authors: | Michael Skinnider [aut, trl, cre], R. Greg Stacey [ctb], Nichollas Scott [ctb], Anders Kristensen [ctb], Leonard Foster [aut, led] |
| Maintainer: | Michael Skinnider <[email protected]> |
| License: | GPL-3 + file LICENSE |
| Version: | 1.23.0 |
| Built: | 2024-11-04 06:08:34 UTC |
| Source: | https://github.com/bioc/PrInCE |
Convert a data frame containing pairwise interactions into an adjacency matrix. The resulting square adjacency matrix contains ones for proteins that are found in interactions and zeroes otherwise.
adjacency_matrix_from_data_frame(dat, symmetric = TRUE, node_columns = c(1, 2))adjacency_matrix_from_data_frame(dat, symmetric = TRUE, node_columns = c(1, 2))
dat |
a data frame containing pairwise interactions |
symmetric |
if true, interactions in both directions will be added to the adjacency matrix |
node_columns |
a vector of length two, denoting either the indices
(integer vector) or column names (character vector) of the columns within
the data frame containing the nodes participating in pairwise interactions;
defaults to the first two columns of the data frame ( |
an adjacency matrix between all interacting proteins
ppi <- data.frame(protein_A = paste0("protein", seq_len(10)), protein_B = paste0("protein", c(rep(3, 2), rep(5, 5), rep(7, 3)))) adj <- adjacency_matrix_from_data_frame(ppi)ppi <- data.frame(protein_A = paste0("protein", seq_len(10)), protein_B = paste0("protein", c(rep(3, 2), rep(5, 5), rep(7, 3)))) adj <- adjacency_matrix_from_data_frame(ppi)
Convert a list of complexes into a pairwise adjacency matrix. The resulting square adjacency matrix contains ones for proteins that are found in the same complex and zeroes otherwise.
adjacency_matrix_from_list(complexes)adjacency_matrix_from_list(complexes)
complexes |
a list of complexes, with each entry containing complex subunits as a character vector |
an adjacency matrix between all complex subunits
data(gold_standard) adj <- adjacency_matrix_from_list(gold_standard)data(gold_standard) adj <- adjacency_matrix_from_list(gold_standard)
Calculate the AIC, corrected AIC, or BIC for a curve fit with a Gaussian mixture model by nonlinear least squares optimization. This function permits the calculation of the AIC/AICc/BIC after rejecting some Gaussians in the model, for example because their centres are outside the bounds of the profile.
gaussian_aic(coefs, chromatogram) gaussian_aicc(coefs, chromatogram) gaussian_bic(coefs, chromatogram)gaussian_aic(coefs, chromatogram) gaussian_aicc(coefs, chromatogram) gaussian_bic(coefs, chromatogram)
coefs |
the coefficients of the Gaussian mixture model, output by
|
chromatogram |
the raw elution profile |
the AIC, corrected AIC, or BIC of the fit model
Identify peaks in co-fractionation profiles by deconvolving peaks in Gaussian mixture models. Models are mixtures of between 1 and 5 Gaussians. Profiles are pre-processed prior to building Gaussians by filtering and cleaning. By default, profiles with fewer than 5 non-missing points, or fewer than 5 consecutive points after imputation of single missing values, are removed. Profiles are cleaned by replacing missing values with near-zero noise, imputing single missing values as the mean of neighboring points, and smoothing with a moving average filter.
build_gaussians( profile_matrix, min_points = 1, min_consecutive = 5, impute_NA = TRUE, smooth = TRUE, smooth_width = 4, max_gaussians = 5, criterion = c("AICc", "AIC", "BIC"), max_iterations = 50, min_R_squared = 0.5, method = c("guess", "random"), filter_gaussians_center = TRUE, filter_gaussians_height = 0.15, filter_gaussians_variance_min = 0.5, filter_gaussians_variance_max = 50 )build_gaussians( profile_matrix, min_points = 1, min_consecutive = 5, impute_NA = TRUE, smooth = TRUE, smooth_width = 4, max_gaussians = 5, criterion = c("AICc", "AIC", "BIC"), max_iterations = 50, min_R_squared = 0.5, method = c("guess", "random"), filter_gaussians_center = TRUE, filter_gaussians_height = 0.15, filter_gaussians_variance_min = 0.5, filter_gaussians_variance_max = 50 )
profile_matrix |
a numeric matrix of co-elution profiles, with proteins
in rows, or a |
min_points |
filter profiles without at least this many total,
non-missing points; passed to |
min_consecutive |
filter profiles without at least this many
consecutive, non-missing points; passed to |
impute_NA |
if true, impute single missing values with the average of
neighboring values; passed to |
smooth |
if true, smooth the chromatogram with a moving average filter;
passed to |
smooth_width |
width of the moving average filter, in fractions;
passed to |
max_gaussians |
the maximum number of Gaussians to fit; defaults to 5.
Note that Gaussian mixtures with more parameters than observed (i.e.,
non-zero or NA) points will not be fit. Passed to
|
criterion |
the criterion to use for model selection;
one of "AICc" (corrected AIC, and default), "AIC", or "BIC". Passed to
|
max_iterations |
the number of times to try fitting the curve with
different initial conditions; defaults to 50. Passed to
|
min_R_squared |
the minimum R-squared value to accept when fitting the
curve with different initial conditions; defaults to 0.5. Passed to
|
method |
the method used to select the initial conditions for
nonlinear least squares optimization (one of "guess" or "random");
see |
filter_gaussians_center |
true or false: filter Gaussians whose centres
fall outside the bounds of the chromatogram. Passed to
|
filter_gaussians_height |
Gaussians whose heights are below this
fraction of the chromatogram height will be filtered. Setting this value to
zero disables height-based filtering of fit Gaussians. Passed to
|
filter_gaussians_variance_min |
Gaussians whose variance falls below
this number of fractions will be filtered. Setting this value to
zero disables filtering. Passed to
|
filter_gaussians_variance_max |
Gaussians whose variance is above
this number of fractions will be filtered. Setting this value to
zero disables filtering. Passed to
|
a list of fit Gaussian mixture models, where each item in the list contains the following five fields: the number of Gaussians used to fit the curve; the R^2 of the fit; the number of iterations used to fit the curve with different initial conditions; the coefficients of the fit model; and the curve predicted by the fit model. Profiles that could not be fit by a Gaussian mixture model above the minimum R-squared cutoff will be absent from the returned list.
data(scott) mat <- clean_profiles(scott[seq_len(5), ]) gauss <- build_gaussians(mat, max_gaussians = 3)data(scott) mat <- clean_profiles(scott[seq_len(5), ]) gauss <- build_gaussians(mat, max_gaussians = 3)
For a given protein, the correlation coefficient to all other proteins in the first condition is calculated, yielding a vector of correlation coefficients. The same procedure is repeated for the second condition, and the two vectors of correlation coefficients are themselves correlated, yielding a metric whereby higher values reflect proteins with unchanging interaction profiles between conditions, while lower values reflect proteins with substantially changing interaction profiles.
calculate_autocorrelation( profile1, profile2, cor_method = c("pearson", "spearman", "kendall"), min_replicates = 1, min_fractions = 1, min_pairs = 0 )calculate_autocorrelation( profile1, profile2, cor_method = c("pearson", "spearman", "kendall"), min_replicates = 1, min_fractions = 1, min_pairs = 0 )
profile1 |
a numeric matrix or data frame with proteins in rows and
fractions in columns, or a |
profile2 |
a numeric matrix or data frame with proteins in rows and
fractions in columns, or a |
cor_method |
the correlation method to use; one of |
min_fractions |
filter proteins not quantified in at least this many fractions |
min_pairs |
remove correlations between protein pairs not co-occuring in at least this many fractions from the autocorrelation calculation |
Note that all of zero, NA, NaN, and infinite values are
all treated equivalently as missing values when applying the
min_fractions and min_pairs filters, but different handling of
missing values will produce different autocorrelation scores.
a named vector of autocorrelation scores for all proteins found in both matrices.
Calculate the six features that are used to discriminate interacting and non-interacting protein pairs based on co-elution profiles in PrInCE, namely: raw Pearson R value, cleaned Pearson R value, raw Pearson P-value, Euclidean distance, co-peak, and co-apex. Optionally, one or more of these can be disabled.
calculate_features( profile_matrix, gaussians, min_pairs = 0, pearson_R_raw = TRUE, pearson_R_cleaned = TRUE, pearson_P = TRUE, euclidean_distance = TRUE, co_peak = TRUE, co_apex = TRUE, n_pairs = FALSE, max_euclidean_quantile = 0.9 )calculate_features( profile_matrix, gaussians, min_pairs = 0, pearson_R_raw = TRUE, pearson_R_cleaned = TRUE, pearson_P = TRUE, euclidean_distance = TRUE, co_peak = TRUE, co_apex = TRUE, n_pairs = FALSE, max_euclidean_quantile = 0.9 )
profile_matrix |
a numeric matrix of co-elution profiles, with proteins
in rows, or a |
gaussians |
a list of Gaussian mixture models fit to the profile matrix
by |
min_pairs |
minimum number of overlapping fractions between any given protein pair to consider a potential interaction |
pearson_R_raw |
if true, include the Pearson correlation (R) between raw profiles as a feature |
pearson_R_cleaned |
if true, include the Pearson correlation (R) between cleaned profiles as a feature |
pearson_P |
if true, include the P-value of the Pearson correlation between raw profiles as a feature |
euclidean_distance |
if true, include the Euclidean distance between cleaned profiles as a feature |
co_peak |
if true, include the 'co-peak score' (that is, the distance, in fractions, between the single highest value of each profile) as a feature |
co_apex |
if true, include the 'co-apex score' (that is, the minimum Euclidean distance between any pair of fit Gaussians) as a feature |
max_euclidean_quantile |
very high Euclidean distance values are trimmed
to avoid numerical precision issues; values above this quantile will be
replaced with the value at this quantile (default: |
a data frame containing the calculated features for all possible protein pairs
Calculate the precision of a list of interactions at each point in the list, given a set of labels.
calculate_precision(labels)calculate_precision(labels)
labels |
a vector of zeroes (FPs) and ones (TPs) |
a vector of the same length giving the precision at each point in the input vector
## calculate features data(scott) data(scott_gaussians) subset <- scott[seq_len(500), ] ## limit to first 500 proteins gauss <- scott_gaussians[names(scott_gaussians) %in% rownames(subset)] features <- calculate_features(subset, gauss) ## make training labels data(gold_standard) ref <- adjacency_matrix_from_list(gold_standard) labels <- make_labels(ref, features) ## predict interactions with naive Bayes classifier ppi <- predict_ensemble(features, labels, classifier = "NB", cv_folds = 3, models = 1) ## tag precision of each interaction ppi$precision <- calculate_precision(ppi$label)## calculate features data(scott) data(scott_gaussians) subset <- scott[seq_len(500), ] ## limit to first 500 proteins gauss <- scott_gaussians[names(scott_gaussians) %in% rownames(subset)] features <- calculate_features(subset, gauss) ## make training labels data(gold_standard) ref <- adjacency_matrix_from_list(gold_standard) labels <- make_labels(ref, features) ## predict interactions with naive Bayes classifier ppi <- predict_ensemble(features, labels, classifier = "NB", cv_folds = 3, models = 1) ## tag precision of each interaction ppi$precision <- calculate_precision(ppi$label)
Test whether an input list of Gaussians conforms to the format expected
by PrInCE: that is, a named list with five fields for each entry, i.e., the
number of Gaussians in the mixture model, the value, the
number of iterations used by nls, the coefficients of
each model, and the fitted curve.
check_gaussians( gaussians, proteins = NULL, replicate_idx = NULL, n_error = 3, pct_warning = 0.1 )check_gaussians( gaussians, proteins = NULL, replicate_idx = NULL, n_error = 3, pct_warning = 0.1 )
gaussians |
the list of Gaussians |
proteins |
the complete set of input proteins |
replicate_idx |
the replicate being analyzed, if input proteins are provided; used to throw more informative error messages |
n_error |
minimum number of proteins that can have fitted Gaussians without throwing an error |
pct_warning |
minimum fraction of proteins that can have fitted Gaussians without giving a warning |
Optionally, some extra checks will be
done on the fraction of proteins in the complete dataset for which a
Gaussian mixture model could be fit, if provided. In particular, the function
will throw an error if fewer than n_error proteins have a fitted
Gaussian, and emit a warning if fewer than pct_warning do.
TRUE if all conditions are met, but throws an error if any
is not
data(scott_gaussians) check_gaussians(scott_gaussians)data(scott_gaussians) check_gaussians(scott_gaussians)
Fit mixtures of one or more Gaussians to the curve formed by a chromatogram profile, and choose the best fitting model using an information criterion of choice.
choose_gaussians( chromatogram, points = NULL, max_gaussians = 5, criterion = c("AICc", "AIC", "BIC"), max_iterations = 10, min_R_squared = 0.5, method = c("guess", "random"), filter_gaussians_center = TRUE, filter_gaussians_height = 0.15, filter_gaussians_variance_min = 0.1, filter_gaussians_variance_max = 50 )choose_gaussians( chromatogram, points = NULL, max_gaussians = 5, criterion = c("AICc", "AIC", "BIC"), max_iterations = 10, min_R_squared = 0.5, method = c("guess", "random"), filter_gaussians_center = TRUE, filter_gaussians_height = 0.15, filter_gaussians_variance_min = 0.1, filter_gaussians_variance_max = 50 )
chromatogram |
a numeric vector corresponding to the chromatogram trace |
points |
optional, the number of non-NA points in the raw data |
max_gaussians |
the maximum number of Gaussians to fit; defaults to 5. Note that Gaussian mixtures with more parameters than observed (i.e., non-zero or NA) points will not be fit. |
criterion |
the criterion to use for model selection; one of "AICc" (corrected AIC, and default), "AIC", or "BIC" |
max_iterations |
the number of times to try fitting the curve with different initial conditions; defaults to 10 |
min_R_squared |
the minimum R-squared value to accept when fitting the curve with different initial conditions; defaults to 0.5 |
method |
the method used to select the initial conditions for
nonlinear least squares optimization (one of "guess" or "random");
see |
filter_gaussians_center |
true or false: filter Gaussians whose centres fall outside the bounds of the chromatogram |
filter_gaussians_height |
Gaussians whose heights are below this fraction of the chromatogram height will be filtered. Setting this value to zero disables height-based filtering of fit Gaussians |
filter_gaussians_variance_min |
Gaussians whose variance is below this threshold will be filtered. Setting this value to zero disables filtering. |
filter_gaussians_variance_max |
Gaussians whose variance is above this threshold will be filtered. Setting this value to zero disables filtering. |
a list with five entries: the number of Gaussians used to fit the curve; the R^2 of the fit; the number of iterations used to fit the curve with different initial conditions; the coefficients of the fit model; and the curve predicted by the fit model.
data(scott) chrom <- clean_profile(scott[1, ]) gauss <- choose_gaussians(chrom, max_gaussians = 3)data(scott) chrom <- clean_profile(scott[1, ]) gauss <- choose_gaussians(chrom, max_gaussians = 3)
Clean a co-elution/co-fractionation profile by (1) imputing single missing values with the average of neighboring values, (2) replacing missing values with random, near-zero noise, and (3) smoothing with a moving average filter.
clean_profile( chromatogram, impute_NA = TRUE, smooth = TRUE, smooth_width = 4, noise_floor = 0.001 )clean_profile( chromatogram, impute_NA = TRUE, smooth = TRUE, smooth_width = 4, noise_floor = 0.001 )
chromatogram |
a numeric vector corresponding to the chromatogram trace |
impute_NA |
if true, impute single missing values with the average of neighboring values |
smooth |
if true, smooth the chromatogram with a moving average filter |
smooth_width |
width of the moving average filter, in fractions |
noise_floor |
mean value of the near-zero noise to add |
a cleaned profile
data(scott) chrom <- scott[16, ] cleaned <- clean_profile(chrom)data(scott) chrom <- scott[16, ] cleaned <- clean_profile(chrom)
Clean a matrix of co-elution/co-fractionation profiles by (1) imputing single missing values with the average of neighboring values, (2) replacing missing values with random, near-zero noise, and (3) smoothing with a moving average filter.
clean_profiles( profile_matrix, impute_NA = TRUE, smooth = TRUE, smooth_width = 4, noise_floor = 0.001 )clean_profiles( profile_matrix, impute_NA = TRUE, smooth = TRUE, smooth_width = 4, noise_floor = 0.001 )
profile_matrix |
a numeric matrix of co-elution profiles, with proteins
in rows, or a |
impute_NA |
if true, impute single missing values with the average of neighboring values |
smooth |
if true, smooth the chromatogram with a moving average filter |
smooth_width |
width of the moving average filter, in fractions |
noise_floor |
mean value of the near-zero noise to add |
a cleaned matrix
data(scott) mat <- scott[c(1, 16), ] mat_clean <- clean_profiles(mat)data(scott) mat <- scott[c(1, 16), ] mat_clean <- clean_profiles(mat)
Calculate the co-apex score for every pair of proteins. This is defined as the minimum Euclidean distance between any two Gaussians fit to each profile.
co_apex(gaussians, proteins = NULL)co_apex(gaussians, proteins = NULL)
gaussians |
a list of Gaussian mixture models fit to the profile matrix
by |
proteins |
all proteins being scored, optionally including those without Gaussian fits |
a matrix of co-apex scores
data(scott_gaussians) gauss <- scott_gaussians[seq_len(25)] CA <- co_apex(gauss)data(scott_gaussians) gauss <- scott_gaussians[seq_len(25)] CA <- co_apex(gauss)
Concatenate features extracted from multiple replicates to a single data frame that will be used as input to a classifier. Doing so allows the classifier to naturally weight evidence for an interaction between each protein pair from each feature in each replicate in proportion to its discriminatory power on known examples.
concatenate_features(feature_list)concatenate_features(feature_list)
feature_list |
a list of feature data frames, as produced by
|
a data frame containing features for all protein pairs across all replicates
Use a permutation testing approach to identify complexes that show a significant tendency to interact, relative to random sets of complexes of equivalent size. The function begins by calculating the Pearson correlation or Euclidean distance between all proteins in the matrix, and
detect_complexes( profile_matrix, complexes, method = c("pearson", "euclidean"), min_pairs = 10, bootstraps = 100, progress = TRUE )detect_complexes( profile_matrix, complexes, method = c("pearson", "euclidean"), min_pairs = 10, bootstraps = 100, progress = TRUE )
profile_matrix |
a matrix of chromatograms, with proteins in the rows
and fractions in the columns, or a |
complexes |
a named list of protein complexes, where the name is the complex name and the entries are proteins within that complex |
method |
method to use to calculate edge weights;
one of |
min_pairs |
the minimum number of pairwise observations to count a correlation or distance towards the z score |
bootstraps |
number of bootstraps to execute to estimate z scores |
progress |
whether to show the progress of the function |
a named vector of z scores for each complex in the input list
data(scott) data(gold_standard) complexes <- gold_standard[lengths(gold_standard) >= 3] z_scores <- detect_complexes(t(scott), complexes) length(na.omit(z_scores)) ## number of complexes that could be tested z_scores[which.max(z_scores)] ## most significant complexdata(scott) data(gold_standard) complexes <- gold_standard[lengths(gold_standard) >= 3] z_scores <- detect_complexes(t(scott), complexes) length(na.omit(z_scores)) ## number of complexes that could be tested z_scores[which.max(z_scores)] ## most significant complex
Filter a matrix of co-elution/co-fractionation profiles by removing profiles without a certain number of non-mising or consecutive points.
filter_profiles(profile_matrix, min_points = 1, min_consecutive = 5)filter_profiles(profile_matrix, min_points = 1, min_consecutive = 5)
profile_matrix |
a numeric matrix of co-elution profiles, with proteins
in rows, or a |
min_points |
filter profiles without at least this many total, non-missing points |
min_consecutive |
filter profiles without at least this many consecutive, non-missing points |
the filtered profile matrix
data(scott) nrow(scott) filtered <- filter_profiles(scott) nrow(scott)data(scott) nrow(scott) filtered <- filter_profiles(scott) nrow(scott)
For a Gaussian mixture model fit to a curve by fit_gaussians,
output the fit curve using the coefficients rather than the nls
object. This allows individual Gaussians to be removed from the fit model:
for example, if their height is below a certain threshold, or their
centres are outside the bounds of the chromatogram.
fit_curve(coef, indices)fit_curve(coef, indices)
coef |
numeric vector of coefficients for a Gaussian mixture model fit
by |
indices |
the indices, or x-values, to predict a fitted curve for (for example, the fractions in a given chromatogram) |
the fitted curve
data(scott) chrom <- clean_profile(scott[1, ]) fit <- fit_gaussians(chrom, n_gaussians = 1) curve <- fit_curve(fit$coefs, seq_along(chrom))data(scott) chrom <- clean_profile(scott[1, ]) fit <- fit_gaussians(chrom, n_gaussians = 1) curve <- fit_curve(fit$coefs, seq_along(chrom))
Fit mixtures of one or more Gaussians to the curve formed by a chromatogram profile, using nonlinear least-squares.
fit_gaussians( chromatogram, n_gaussians, max_iterations = 10, min_R_squared = 0.5, method = c("guess", "random"), filter_gaussians_center = TRUE, filter_gaussians_height = 0.15, filter_gaussians_variance_min = 0.1, filter_gaussians_variance_max = 50 )fit_gaussians( chromatogram, n_gaussians, max_iterations = 10, min_R_squared = 0.5, method = c("guess", "random"), filter_gaussians_center = TRUE, filter_gaussians_height = 0.15, filter_gaussians_variance_min = 0.1, filter_gaussians_variance_max = 50 )
chromatogram |
a numeric vector corresponding to the chromatogram trace |
n_gaussians |
the number of Gaussians to fit |
max_iterations |
the number of times to try fitting the curve with different initial conditions; defaults to 10 |
min_R_squared |
the minimum R-squared value to accept when fitting the curve with different initial conditions; defaults to 0.5 |
method |
the method used to select the initial conditions for
nonlinear least squares optimization (one of "guess" or "random");
see |
filter_gaussians_center |
true or false: filter Gaussians whose centres fall outside the bounds of the chromatogram |
filter_gaussians_height |
Gaussians whose heights are below this fraction of the chromatogram height will be filtered. Setting this value to zero disables height-based filtering of fit Gaussians |
filter_gaussians_variance_min |
Gaussians whose variance falls below this number of fractions will be filtered. Setting this value to zero disables filtering. |
filter_gaussians_variance_max |
Gaussians whose variance is above this number of fractions will be filtered. Setting this value to zero disables filtering. |
a list with six entries: the number of Gaussians used to fit the curve; the R^2 of the fit; the number of iterations used to fit the curve with different initial conditions; the coefficients of the fit model; and the fit curve predicted by the fit model.
data(scott) chrom <- clean_profile(scott[1, ]) fit <- fit_gaussians(chrom, n_gaussians = 1)data(scott) chrom <- clean_profile(scott[1, ]) fit <- fit_gaussians(chrom, n_gaussians = 1)
A reference set of 467 experimentally confirmed human protein complexes, derived from the EBI Complex Portal database.
data(gold_standard)data(gold_standard)
a list containing 467 entries (character vectors)
467 protein complexes, ranging in size from 2 to 44 proteins and involving 877 proteins in total, to provide a reference set of true positive and true negative interactions (intra- and inter-complex interactions, respectively) for demonstration in PrInCE analysis of a co-elution dataset. Other "gold standards" are possible in practice, most notably the CORUM database; however, the Complex Portal reference set is included in this package due to its CC-BY license.
https://www.ebi.ac.uk/complexportal/complex/organisms
Impute single missing values within a chromatogram profile as the average of their neighbors.
impute_neighbors(chromatogram)impute_neighbors(chromatogram)
chromatogram |
a numeric vector corresponding to the chromatogram trace |
the imputed chromatogram
data(scott) chrom <- scott[16, ] imputed <- impute_neighbors(chrom)data(scott) chrom <- scott[16, ] imputed <- impute_neighbors(chrom)
Test whether a network is unweighted
is_unweighted(network)is_unweighted(network)
network |
the network to analyze |
true if the input network is a square logical or numeric matrix
data(gold_standard) adj <- adjacency_matrix_from_list(gold_standard) is_unweighted(adj) ## returns TRUEdata(gold_standard) adj <- adjacency_matrix_from_list(gold_standard) is_unweighted(adj) ## returns TRUE
Test whether a network is weighted
is_weighted(network)is_weighted(network)
network |
the network to analyze |
true if the input network is a square numeric matrix with more than two values
data(gold_standard) adj <- adjacency_matrix_from_list(gold_standard) is_weighted(adj) ## returns FALSEdata(gold_standard) adj <- adjacency_matrix_from_list(gold_standard) is_weighted(adj) ## returns FALSE
Co-elution profiles derived from size exclusion chromatography (SEC) of HeLa cell lysates.
data(kristensen)data(kristensen)
a data frame with 1875 rows and 48 columns, with proteins in rows and SEC fractions in columns
Protein quantitation was accomplished by SILAC (stable isotopic labelling
by amino acids in cell culture), and is ratiometric, i.e., it reflects
the ratio between the intensity of the heavy isotope and the light isotope
("H/L"). The dataset was initially described in Kristensen et al.,
Nat. Methods 2012. The medium isotope channel from replicate
1 (Supplementary Table 1a in the online supplementary information) is
included in the PrInCE package. The R script used to generate
this matrix from the supplementary materials of the paper is provided in the
data-raw directory of the package source code.
https://www.nature.com/articles/nmeth.2131
kristensen datasetThe kristensen dataset consists of protein
co-migration profiles derived from size exclusion chromatography (SEC) of
unstimulated HeLa cell lysates.
The kristensen_gaussians object contains Gaussian mixture models fit by
the function build_gaussians; this is bundled with the
R package in order to expedite the demonstration code, as the process of
Gaussian fitting is one of the more time-consuming aspects of the package.
data(kristensen_gaussians)data(kristensen_gaussians)
a named list with 1117 entries; names are proteins, and list items conain information about fitted Gaussians in the format that PrInCE expects
As with the kristensen dataset, the code used to
generate this data object is provided in the data-raw directory of the
package source.
Convert a data frame containing pairwise interactions, and a score or other data associated with each interaction, into a feature vector that matches the dimensions of a data frame used as input to a classifier, such as a naive Bayes, random forests, or support vector machine classifier.
make_feature_from_data_frame( dat, target, dat_node_cols = c(1, 2), target_node_cols = c(1, 2), feature_col = 3, default_value = NA )make_feature_from_data_frame( dat, target, dat_node_cols = c(1, 2), target_node_cols = c(1, 2), feature_col = 3, default_value = NA )
dat |
a data frame containing pairwise interactions and a feature to be converted to a vector in a third column |
target |
the data frame of features that will be provided as input to a classifier |
dat_node_cols |
a vector of length two, denoting either the indices
(integer vector) or column names (character vector) of the columns within
the feature data frame; defaults to the first two columns of the data
frame ( |
target_node_cols |
a vector of length two, denoting either the indices
(integer vector) or column names (character vector) of the columns within
the target data frame; defaults to the first two columns of the data
frame ( |
feature_col |
the name or index of the column in the first data frame that contains a feature for each interaction |
default_value |
the default value for protein pairs that are not in the
first data frame (set, by default, to |
a vector matching the dimensions and order of the feature data frame, to use as input for a classifier in interaction prediction
Convert a gene or protein expression matrix into a feature vector that matches the dimensions of a data frame used as input to a classifier, such as a naive Bayes, random forests, or support vector machine classifier, by calculating the correlation between each pair of genes or proteins.
make_feature_from_expression(expr, dat, node_columns = c(1, 2), ...)make_feature_from_expression(expr, dat, node_columns = c(1, 2), ...)
expr |
a matrix containing gene or protein expression data, with genes/proteins in columns and samples in rows |
dat |
the data frame of features to be used by the classifier,
with protein pairs in the columns specified by the |
node_columns |
a vector of length two, denoting either the indices
(integer vector) or column names (character vector) of the columns within
the data frame containing the nodes participating in pairwise interactions;
defaults to the first two columns of the data frame ( |
... |
arguments passed to |
a vector matching the dimensions and order of the feature data frame, to use as input for a classifier in interaction prediction
Construct a set of initial conditions for curve fitting using nonlinear least squares using a mixture of Gaussians. The "guess" method ports code from the Matlab release of PrInCE. This method finds local maxima within the chromatogram, orders them by their separation (in number of fractions) from the previous local maxima, and uses the positions and heights of these local maxima (+/- some random noise) as initial conditions for Gaussian curve-fitting. The "random" method simply picks random values within the fraction and intensity intervals as starting points for Gaussian curve-fitting. The initial value of sigma is set by default to a random number within +/- 0.5 of two for both modes; this is based on our manual inspection of a large number of chromatograms.
make_initial_conditions( chromatogram, n_gaussians, method = c("guess", "random"), sigma_default = 2, sigma_noise = 0.5, mu_noise = 1.5, A_noise = 0.5 )make_initial_conditions( chromatogram, n_gaussians, method = c("guess", "random"), sigma_default = 2, sigma_noise = 0.5, mu_noise = 1.5, A_noise = 0.5 )
chromatogram |
a numeric vector corresponding to the chromatogram trace |
n_gaussians |
the number of Gaussians being fit |
method |
one of "guess" or "random", discussed above |
sigma_default |
the default mean initial value of sigma |
sigma_noise |
the amount of random noise to add or subtract from the default mean initial value of sigma |
mu_noise |
the amount of random noise to add or subtract from the Gaussian centers in "guess" mode |
A_noise |
the amount of random noise to add or subtract from the Gaussian heights in "guess" mode |
a list of three numeric vectors (A, mu, and sigma), each having a length equal to the maximum number of Gaussians to fit
data(scott) chrom <- clean_profile(scott[16, ]) set.seed(0) start <- make_initial_conditions(chrom, n_gaussians = 2, method = "guess")data(scott) chrom <- clean_profile(scott[16, ]) set.seed(0) start <- make_initial_conditions(chrom, n_gaussians = 2, method = "guess")
Create labels for a classifier for protein pairs in the same order as in a dataset that will be used as input to a classifier, in a memory-friendly way.
make_labels(gold_standard, dat, node_columns = c(1, 2), protein_groups = NULL)make_labels(gold_standard, dat, node_columns = c(1, 2), protein_groups = NULL)
gold_standard |
an adjacency matrix of gold-standard interactions |
dat |
a data frame with interacting proteins in the first two columns |
node_columns |
a vector of length two, denoting either the indices
(integer vector) or column names (character vector) of the columns within
the data frame containing the nodes participating in pairwise interactions;
defaults to the first two columns of the data frame ( |
protein_groups |
optionally, specify a list linking each protein in the first two columns of the input data frame to a protein group |
a vector of the same length as the input dataset, containing
NAs for protein pairs not in the gold standard and ones or zeroes
based on the content of the adjacency matrix
data(gold_standard) adj <- adjacency_matrix_from_list(gold_standard) proteins <- unique(unlist(gold_standard)) dat <- data.frame(protein_A = sample(proteins, 10), protein_B = sample(proteins, 10)) labels <- make_labels(adj, dat)data(gold_standard) adj <- adjacency_matrix_from_list(gold_standard) proteins <- unique(unlist(gold_standard)) dat <- data.frame(protein_A = sample(proteins, 10), protein_B = sample(proteins, 10)) labels <- make_labels(adj, dat)
Match the row and column names of a square feature matrix to the row names
of a profile matrix, adding rows/columns containing NAs when
proteins in the profile matrix are missing from the feature matrix.
match_matrix_dimensions(query, profile_matrix)match_matrix_dimensions(query, profile_matrix)
query |
a square matrix containing features for pairs of proteins |
profile_matrix |
the profile matrix for which interactions are being predicted |
a square matrix with the same row and column names as the input profile matrix, for use in interaction prediction
data(gold_standard) subset <- adjacency_matrix_from_list(gold_standard[seq(1, 200)]) target <- adjacency_matrix_from_list(gold_standard) matched <- match_matrix_dimensions(subset, target) dim(subset) dim(target) dim(matched)data(gold_standard) subset <- adjacency_matrix_from_list(gold_standard[seq(1, 200)]) target <- adjacency_matrix_from_list(gold_standard) matched <- match_matrix_dimensions(subset, target) dim(subset) dim(target) dim(matched)
Use an ensemble of classifiers to predict interactions from co-elution dataset features. The ensemble approach ensures that results are robust to the partitioning of the dataset into folds. For each model, the median of classifier scores across all folds is calculated. Then, the median of all such medians across all models is calculated.
predict_ensemble( dat, labels, classifier = c("NB", "SVM", "RF", "LR"), models = 1, cv_folds = 10, trees = 500, node_columns = c(1, 2) )predict_ensemble( dat, labels, classifier = c("NB", "SVM", "RF", "LR"), models = 1, cv_folds = 10, trees = 500, node_columns = c(1, 2) )
dat |
a data frame containing interacting gene/protein pairs in the first two columns, and the features to use for classification in the remaining columns |
labels |
labels for each interaction in |
classifier |
the type of classifier to use; one of |
models |
the number of classifiers to train |
cv_folds |
the number of folds to split the reference dataset into when training each classifier. By default, each classifier uses ten-fold cross-validation, i.e., the classifier is trained on 90% of the dataset and used to classify the remaining 10% |
trees |
for random forest classifiers only, the number of trees to grow for each fold |
node_columns |
a vector of length two, denoting either the indices
(integer vector) or column names (character vector) of the columns within
the input data frame containing the nodes participating in pairwise
interactions; defaults to the first two columns of the data frame
( |
the input data frame of pairwise interactions, ranked by the median of classifier scores across all ensembled models
## calculate features data(scott) data(scott_gaussians) subset <- scott[seq_len(500), ] ## limit to first 500 proteins gauss <- scott_gaussians[names(scott_gaussians) %in% rownames(subset)] features <- calculate_features(subset, gauss) ## make training labels data(gold_standard) ref <- adjacency_matrix_from_list(gold_standard) labels <- make_labels(ref, features) ## predict interactions with naive Bayes classifier ppi <- predict_ensemble(features, labels, classifier = "NB", cv_folds = 3, models = 1)## calculate features data(scott) data(scott_gaussians) subset <- scott[seq_len(500), ] ## limit to first 500 proteins gauss <- scott_gaussians[names(scott_gaussians) %in% rownames(subset)] features <- calculate_features(subset, gauss) ## make training labels data(gold_standard) ref <- adjacency_matrix_from_list(gold_standard) labels <- make_labels(ref, features) ## predict interactions with naive Bayes classifier ppi <- predict_ensemble(features, labels, classifier = "NB", cv_folds = 3, models = 1)
Discriminate interacting from non-interacting protein pairs by training a
machine learning model on a set of labelled examples, given a set of
features derived from a co-elution profile matrix (see
calculate_features.
predict_interactions( features, gold_standard, classifier = c("NB", "SVM", "RF", "LR", "ensemble"), verbose = FALSE, models = 10, cv_folds = 10, trees = 500 )predict_interactions( features, gold_standard, classifier = c("NB", "SVM", "RF", "LR", "ensemble"), verbose = FALSE, models = 10, cv_folds = 10, trees = 500 )
features |
a data frame with proteins in the first two columns, and features to be passed to the classifier in the remaining columns |
gold_standard |
an adjacency matrix of "gold standard" interactions used to train the classifier |
classifier |
the type of classifier to use: one of |
verbose |
if |
models |
the number of classifiers to train and average across, each with a different k-fold cross-validation split |
cv_folds |
the number of folds to use for k-fold cross-validation |
trees |
for random forests only, the number of trees in the forest |
PrInCE implements four different classifiers (naive Bayes, support vector machine, random forest, and logistic regression). Naive Bayes is used as a default. The classifiers are trained on the gold standards using a ten-fold cross-validation procedure, training on 90 that are part of the training data, the held-out split is used to assign a classifier score, whereas for the remaining protein pairs, the median of all ten folds is used. Furthermore, to ensure the results are not sensitive to the precise classifier split used, an ensemble of multiple classifiers (ten, by default) is trained, and the classifier score is subsequently averaged across classifiers.
PrInCE can also ensemble across multiple different types of classifiers,
by supplying the "ensemble" option to the classifier argument.
a ranked data frame of pairwise interactions, with the classifier score, label, and cumulative precision for each interaction
## calculate features data(scott) data(scott_gaussians) subset <- scott[seq_len(500), ] ## limit to first 500 proteins gauss <- scott_gaussians[names(scott_gaussians) %in% rownames(subset)] features <- calculate_features(subset, gauss) ## load training data data(gold_standard) ref <- adjacency_matrix_from_list(gold_standard) ## predict interactions ppi <- predict_interactions(features, ref, cv_folds = 3, models = 1)## calculate features data(scott) data(scott_gaussians) subset <- scott[seq_len(500), ] ## limit to first 500 proteins gauss <- scott_gaussians[names(scott_gaussians) %in% rownames(subset)] features <- calculate_features(subset, gauss) ## load training data data(gold_standard) ref <- adjacency_matrix_from_list(gold_standard) ## predict interactions ppi <- predict_interactions(features, ref, cv_folds = 3, models = 1)
PrInCE is a computational approach to infer protein-protein interaction networks from co-elution proteomics data, also called co-migration, co-fractionation, or protein correlation profiling. This family of methods separates interacting protein complexes on the basis of their diameter or biochemical properties. Protein-protein interactions can then be inferred for pairs of proteins with similar elution profiles. PrInCE implements a machine-learning approach to identify protein-protein interactions given a set of labelled examples, using features derived exclusively from the data. This allows PrInCE to infer high-quality protein interaction networks from raw proteomics data, without bias towards known interactions or functionally associated proteins, making PrInCE a unique resource for discovery.
PrInCE( profiles, gold_standard, gaussians = NULL, precision = NULL, verbose = FALSE, min_points = 1, min_consecutive = 5, min_pairs = 3, impute_NA = TRUE, smooth = TRUE, smooth_width = 4, max_gaussians = 5, max_iterations = 50, min_R_squared = 0.5, method = c("guess", "random"), criterion = c("AICc", "AIC", "BIC"), pearson_R_raw = TRUE, pearson_R_cleaned = TRUE, pearson_P = TRUE, euclidean_distance = TRUE, co_peak = TRUE, co_apex = TRUE, n_pairs = FALSE, classifier = c("NB", "SVM", "RF", "LR", "ensemble"), models = 1, cv_folds = 10, trees = 500 )PrInCE( profiles, gold_standard, gaussians = NULL, precision = NULL, verbose = FALSE, min_points = 1, min_consecutive = 5, min_pairs = 3, impute_NA = TRUE, smooth = TRUE, smooth_width = 4, max_gaussians = 5, max_iterations = 50, min_R_squared = 0.5, method = c("guess", "random"), criterion = c("AICc", "AIC", "BIC"), pearson_R_raw = TRUE, pearson_R_cleaned = TRUE, pearson_P = TRUE, euclidean_distance = TRUE, co_peak = TRUE, co_apex = TRUE, n_pairs = FALSE, classifier = c("NB", "SVM", "RF", "LR", "ensemble"), models = 1, cv_folds = 10, trees = 500 )
profiles |
the co-elution profile matrix, or a list of profile matrices
if replicate experiments were performed. Can be a single numeric matrix,
with proteins in rows and fractions in columns, or a list of matrices.
Alternatively, can be provided as a single
|
gold_standard |
a set of 'gold standard' interactions, used to train the classifier. Can be provided either as an adjacency matrix, in which both rows and columns correspond to protein IDs in the co-elution matrix or matrices, or as a list of proteins in the same complex, which will be converted to an adjacency matrix by PrInCE. Zeroes in the adjacency matrix are interpreted by PrInCE as "true negatives" when calculating precision. |
gaussians |
optionally, provide Gaussian mixture models fit by
the |
precision |
optionally, return only interactions above the given
precision; by default, all interactions are returned and the user can
subsequently threshold the list using the
|
verbose |
if |
min_points |
filter profiles without at least this many total,
non-missing points; passed to |
min_consecutive |
filter profiles without at least this many
consecutive, non-missing points; passed to |
min_pairs |
minimum number of overlapping fractions between any given protein pair to consider a potential interaction |
impute_NA |
if true, impute single missing values with the average of
neighboring values; passed to |
smooth |
if true, smooth the chromatogram with a moving average filter;
passed to |
smooth_width |
width of the moving average filter, in fractions;
passed to |
max_gaussians |
the maximum number of Gaussians to fit; defaults to 5.
Note that Gaussian mixtures with more parameters than observed (i.e.,
non-zero or NA) points will not be fit. Passed to
|
max_iterations |
the number of times to try fitting the curve with
different initial conditions; defaults to 50. Passed to
|
min_R_squared |
the minimum R-squared value to accept when fitting the
curve with different initial conditions; defaults to 0.5. Passed to
|
method |
the method used to select the initial conditions for
nonlinear least squares optimization (one of "guess" or "random");
see |
criterion |
the criterion to use for model selection;
one of "AICc" (corrected AIC, and default), "AIC", or "BIC". Passed to
|
pearson_R_raw |
if true, include the Pearson correlation (R) between raw profiles as a feature |
pearson_R_cleaned |
if true, include the Pearson correlation (R) between cleaned profiles as a feature |
pearson_P |
if true, include the P-value of the Pearson correlation between raw profiles as a feature |
euclidean_distance |
if true, include the Euclidean distance between cleaned profiles as a feature |
co_peak |
if true, include the 'co-peak score' (that is, the distance, in fractions, between the single highest value of each profile) as a feature |
co_apex |
if true, include the 'co-apex score' (that is, the minimum Euclidean distance between any pair of fit Gaussians) as a feature |
n_pairs |
if |
classifier |
the type of classifier to use: one of |
models |
the number of classifiers to train and average across, each with a different k-fold cross-validation split |
cv_folds |
the number of folds to use for k-fold cross-validation |
trees |
for random forests only, the number of trees in the forest |
PrInCE takes as input a co-elution matrix, with detected proteins in rows and
fractions as columns, and a set of 'gold standard' true positives and true
negatives. If replicate experiments were performed, a list of co-elution
matrices can be provided as input. PrInCE will construct features for each
replicate separately and use features from all replicates as input to the
classifier. The 'gold standard' can be either a data frame or adjacency
matrix of known interactions (and non-interactions), or a list of protein
complexes. For computational convenience, Gaussian mixture models can be
pre-fit to every profile and provided separately to the PrInCE
function. The matrix, or matrices, can be provided to PrInCE either as
numeric matrices or as MSnSet objects.
PrInCE implements three different types of classifiers to predict protein-protein interaction networks, including naive Bayes (the default), random forests, and support vector machines. The classifiers are trained on the gold standards using a ten-fold cross-validation procedure, training on 90 that are part of the training data, the held-out split is used to assign a classifier score, whereas for the remaining protein pairs, the median of all ten folds is used. Furthermore, to ensure the results are not sensitive to the precise classifier split used, an ensemble of multiple classifiers (ten, by default) is trained, and the classifier score is subsequently averaged across classifiers. PrInCE can also ensemble across a set of classifiers.
By default, PrInCE calculates six features from each pair of co-elution profiles as input to the classifier, including conventional similarity metrics but also several features specifically adapted to co-elution proteomics. For example, one such feature is derived from fitting a Gaussian mixture model to each elution profile, then calculating the smallest Euclidean distance between any pair of fitted Gaussians. The complete set of features includes:
the Pearson correlation between raw co-elution profiles;
the p-value of the Pearson correlation between raw co-elution profiles;
the Pearson correlation between cleaned profiles, which are generated
by imputing single missing values with the mean of their neighbors,
replacing remaining missing values with random near-zero noise, and
smoothing the profiles using a moving average filter (see
clean_profile);
the Euclidean distance between cleaned profiles;
the 'co-peak' score, defined as the distance, in fractions, between the maximum values of each profile; and
the 'co-apex' score, defined as the minimum Euclidean distance between any pair of fit Gaussians
The output of PrInCE is a ranked data frame, containing the classifier score for every possible protein pair. PrInCE also calculates the precision at every point in this ranked list, using the 'gold standard' set of protein complexes or binary interactions. Our recommendation is to select a threshold for the precision and use this to construct an unweighted protein interaction network.
a ranked data frame of interacting proteins, with the precision at each point in the list
Stacey RG, Skinnider MA, Scott NE, Foster LJ (2017). “A rapid and accurate approach for prediction of interactomes from co-elution data (PrInCE).” BMC Bioinformatics, 18(1), 457.
Scott NE, Brown LM, Kristensen AR, Foster LJ (2015). “Development of a computational framework for the analysis of protein correlation profiling and spatial proteomics experiments.” Journal of Proteomics, 118, 112–129.
Kristensen AR, Gsponer J, Foster LJ (2012). “A high-throughput approach for measuring temporal changes in the interactome.” Nature Methods, 9(9), 907–909.
Skinnider MA, Stacey RG, Foster LJ (2018). “Genomic data integration systematically biases interactome mapping.” PLoS Computational Biology, 14(10), e1006474.
data(scott) data(scott_gaussians) data(gold_standard) # analyze only the first 100 profiles subset <- scott[seq_len(500), ] gauss <- scott_gaussians[names(scott_gaussians) %in% rownames(subset)] ppi <- PrInCE(subset, gold_standard, gaussians = gauss, models = 1, cv_folds = 3 )data(scott) data(scott_gaussians) data(gold_standard) # analyze only the first 100 profiles subset <- scott[seq_len(500), ] gauss <- scott_gaussians[names(scott_gaussians) %in% rownames(subset)] ppi <- PrInCE(subset, gold_standard, gaussians = gauss, models = 1, cv_folds = 3 )
Replace missing data within each numeric column of a data frame with the column median, plus or minus some random noise, in order to train classifiers that do not easily ignore missing data (e.g. random forests or support vector machines).
replace_missing_data(dat, noise_pct = 0.05)replace_missing_data(dat, noise_pct = 0.05)
dat |
the data frame to replace missing data in |
noise_pct |
the standard deviation of the random normal distribution from which to draw added noise, expressed as a percentage of the standard deviation of the non-missing values in each column |
a data frame with missing values in each numeric column replaced by the column median, plus or minus some random noise
Co-elution profiles derived from size exclusion chromatography (SEC) of cytoplasmic fractions from Jurkat T cells, 4 hours following Fas stimulation.
data(scott)data(scott)
a data frame with 1560 rows and 55 columns, with proteins in rows and SEC fractions in columns
Protein quantitation was accomplished by SILAC (stable isotopic labelling
by amino acids in cell culture), and is ratiometric, i.e., it reflects
the ratio between the intensity of the heavy isotope and the light isotope
("H/L"). The dataset was initially described in Scott et al.,
Mol. Syst. Biol. 2017. The heavy isotope channel from replicate
1 is included in the PrInCE package. The R script used to generate
this matrix from the supplementary materials of the paper is provided in the
data-raw directory of the package source code.
http://msb.embopress.org/content/13/1/906
scott datasetThe scott dataset consists of protein
co-migration profiles derived from size exclusion chromatography (SEC) of
cytoplasmic fractions from Jurkat T cells, 4 hours following Fas stimulation.
The scott_gaussians object contains Gaussian mixture models fit by
the function build_gaussians; this is bundled with the
R package in order to expedite the demonstration code, as the process of
Gaussian fitting is one of the more time-consuming aspects of the package.
data(scott_gaussians)data(scott_gaussians)
a named list with 970 entries; names are proteins, and list items conain information about fitted Gaussians in the format that PrInCE expects
As with the scott dataset, the code used to generate
this data object is provided in the data-raw directory of the package
source.
Threshold interactions at a given precision cutoff
threshold_precision(interactions, threshold)threshold_precision(interactions, threshold)
interactions |
the ranked list of interactions output by
|
threshold |
the minimum precision of the unweighted interaction network to return |
the subset of the original ranked list at the given precision
data(scott) data(scott_gaussians) data(gold_standard) # analyze only the first 100 profiles subset <- scott[seq_len(500), ] gauss <- scott_gaussians[names(scott_gaussians) %in% rownames(subset)] ppi <- PrInCE(subset, gold_standard, gaussians = gauss, models = 1, cv_folds = 3 ) network <- threshold_precision(ppi, threshold = 0.5) nrow(network)data(scott) data(scott_gaussians) data(gold_standard) # analyze only the first 100 profiles subset <- scott[seq_len(500), ] gauss <- scott_gaussians[names(scott_gaussians) %in% rownames(subset)] ppi <- PrInCE(subset, gold_standard, gaussians = gauss, models = 1, cv_folds = 3 ) network <- threshold_precision(ppi, threshold = 0.5) nrow(network)