skbio.stats.composition.ancom#
- skbio.stats.composition.ancom(table, grouping, alpha=0.05, tau=0.02, theta=0.1, p_adjust='holm', significance_test='f_oneway', percentiles=(0.0, 25.0, 50.0, 75.0, 100.0), multiple_comparisons_correction='holm-bonferroni')[source]#
Perform a differential abundance test using ANCOM.
Analysis of composition of microbiomes (ANCOM) is done by calculating pairwise log ratios between all features and performing a significance test to determine if there is a significant difference in feature ratios with respect to the variable of interest.
In an experiment with only two treatments, this tests the following hypothesis for feature \(i\):
\[H_{0i}: \mathbb{E}[\ln(u_i^{(1)})] = \mathbb{E}[\ln(u_i^{(2)})]\]where \(u_i^{(1)}\) is the mean abundance for feature \(i\) in the first group and \(u_i^{(2)}\) is the mean abundance for feature \(i\) in the second group.
- Parameters:
- tablepd.DataFrame
A 2-D matrix of strictly positive values (i.e. counts or proportions) where the rows correspond to samples and the columns correspond to features.
- groupingpd.Series
Vector indicating the assignment of samples to groups. For example, these could be strings or integers denoting which group a sample belongs to. It must be the same length as the samples in table. The index must be the same on table and grouping but need not be in the same order.
- alphafloat, optional
Significance level for each of the statistical tests. This can can be anywhere between 0 and 1 exclusive.
- taufloat, optional
A constant used to determine an appropriate cutoff. A value close to zero indicates a conservative cutoff. This can can be anywhere between 0 and 1 exclusive.
- thetafloat, optional
Lower bound for the proportion for the W-statistic. If all W- statistics are lower than theta, then no features will be detected to be significantly different. This can can be anywhere between 0 and 1 exclusive.
- p_adjuststr or None, optional
Method to correct p-values for multiple comparisons. Options are Holm- Boniferroni (“holm” or “holm-bonferroni”) (default), Benjamini- Hochberg (“bh”, “fdr_bh” or “benjamini-hochberg”), or any method supported by statsmodels’
multipletests
function. Case-insensitive. If None, no correction will be performed.Changed in version 0.6.0: Replaces
multiple_comparisons_correction
for conciseness.- significance_teststr or callable, optional
A function to test for significance between classes. It must be able to accept at least two vectors of floats and returns a test statistic and a p-value. Functions under
scipy.stats
can be directly specified by name. The default is one-way ANOVA (“f_oneway”).Changed in version 0.6.0: Accepts test names in addition to functions.
- percentilesiterable of floats, optional
Percentile abundances to return for each feature in each group. By default, will return the minimum, 25th percentile, median, 75th percentile, and maximum abundances for each feature in each group.
- multiple_comparisons_correctionstr or None, optional
Alias for
p_adjust
. For backward compatibility. Deprecated.
- Returns:
- pd.DataFrame
A table of features, their W-statistics and whether the null hypothesis is rejected.
W
: W-statistic, or the number of features that the current feature is tested to be significantly different against.Reject null hypothesis
: Whether the feature is differentially abundant across groups (True
) or not (False
).
- pd.DataFrame
A table of features and their percentile abundances in each group. If
percentiles
is empty, this will be an emptypd.DataFrame
. The rows in this object will be features, and the columns will be a multi-index where the first index is the percentile, and the second index is the group.
Warning
multiple_comparisons_correction
is deprecated as of0.6.0
. It has been renamed top_adjust
.significance_test=None
is deprecated as of0.6.0
. The default value is now “f_oneway”.See also
Notes
The developers of ANCOM recommend the following significance tests ([1], Supplementary File 1, top of page 11):
If there are two groups, use the standard parametric t-test (
ttest_ind
) or the non-parametric Mann-Whitney rank test (mannwhitneyu
).For paired samples, use the parametric paired t-test (
ttest_rel
) or the non-parametric Wilcoxon signed-rank test (wilcoxon
).If there are more than two groups, use the parametric one-way ANOVA (
f_oneway
) or the non-parametric Kruskal-Wallis test (kruskal
).If there are multiple measurements obtained from the individuals, use a Friedman test (
friedmanchisquare
).
Because one-way ANOVA is equivalent to the standard t-test when the number of groups is two, we default to
f_oneway
here, which can be used when there are two or more groups.Users should refer to the documentation of these tests in SciPy to understand the assumptions made by each test.
This method cannot handle any zero counts as input, since the logarithm of zero cannot be computed. While this is an unsolved problem, many studies, including [1], have shown promising results by adding pseudocounts to all values in the matrix. In [1], a pseudocount of 0.001 was used, though the authors note that a pseudocount of 1.0 may also be useful. Zero counts can also be addressed using the
multi_replace
method.References
Examples
>>> from skbio.stats.composition import ancom >>> import pandas as pd
Let’s load in a DataFrame with six samples and seven features (e.g., these may be bacterial taxa):
>>> table = pd.DataFrame([[12, 11, 10, 10, 10, 10, 10], ... [9, 11, 12, 10, 10, 10, 10], ... [1, 11, 10, 11, 10, 5, 9], ... [22, 21, 9, 10, 10, 10, 10], ... [20, 22, 10, 10, 13, 10, 10], ... [23, 21, 14, 10, 10, 10, 10]], ... index=['s1', 's2', 's3', 's4', 's5', 's6'], ... columns=['b1', 'b2', 'b3', 'b4', 'b5', 'b6', ... 'b7'])
Then create a grouping vector. In this example, there is a treatment group and a placebo group.
>>> grouping = pd.Series(['treatment', 'treatment', 'treatment', ... 'placebo', 'placebo', 'placebo'], ... index=['s1', 's2', 's3', 's4', 's5', 's6'])
Now run
ancom
to determine if there are any features that are significantly different in abundance between the treatment and the placebo groups. The first DataFrame that is returned contains the ANCOM test results, and the second contains the percentile abundance data for each feature in each group.>>> ancom_df, percentile_df = ancom(table, grouping) >>> ancom_df['W'] b1 0 b2 4 b3 0 b4 1 b5 1 b6 0 b7 1 Name: W, dtype: ...
The W-statistic is the number of features that a single feature is tested to be significantly different against. In this scenario,
b2
was detected to have significantly different abundances compared to four of the other features. To summarize the results from the W-statistic, let’s take a look at the results from the hypothesis test. TheReject null hypothesis
column in the table indicates whether the null hypothesis was rejected, and that a feature was therefore observed to be differentially abundant across the groups.>>> ancom_df['Reject null hypothesis'] b1 False b2 True b3 False b4 False b5 False b6 False b7 False Name: Reject null hypothesis, dtype: bool
From this we can conclude that only
b2
was significantly different in abundance between the treatment and the placebo. We still don’t know, for example, in which groupb2
was more abundant. We therefore may next be interested in comparing the abundance ofb2
across the two groups. We can do that using the second DataFrame that was returned. Here we compare the median (50th percentile) abundance ofb2
in the treatment and placebo groups:>>> percentile_df[50.0].loc['b2'] Group placebo 21.0 treatment 11.0 Name: b2, dtype: float64
We can also look at a full five-number summary for
b2
in the treatment and placebo groups:>>> percentile_df.loc['b2'] Percentile Group 0.0 placebo 21.0 25.0 placebo 21.0 50.0 placebo 21.0 75.0 placebo 21.5 100.0 placebo 22.0 0.0 treatment 11.0 25.0 treatment 11.0 50.0 treatment 11.0 75.0 treatment 11.0 100.0 treatment 11.0 Name: b2, dtype: float64
Taken together, these data tell us that
b2
is present in significantly higher abundance in the placebo group samples than in the treatment group samples.