Estimate the cross-talk rate de novo (Edgar 2018)

Estimate the cross-talk rate f used by uncross2_pq() directly from the OTU table, without control/mock samples, following the de novo procedure of Edgar (2018, doi:10.1101/400762 ).

For each OTU i with total abundance N_i, a count c is deemed low (i.e. consistent with cross-talk) when 0 < c <= s*N_i/n, where n is the number of samples (default s = 0.1, so "ten times below the mean"). Let S_i be the number of low counts and X_i their sum. An OTU is a candidate when N_i >= nmin and S_i >= smin, and its rate is

rate = "edgar2018" (paper eq. 2): f_i = (X_i/N_i)*(n/S_i)
rate = "fraction": f_i = X_i/N_i, the raw low-read fraction, which is what the usearch otutab_xtalk report prints as each OTU's rate.

Candidates with f_i > fmax are rejected. If fewer than cmin candidates remain the estimate is undetermined and NA is returned (a real UNCROSS2 outcome; several datasets in Edgar's Table 5 are undetermined). Otherwise the returned rate is the median f_i over candidates.

Usage

estimate_xtalk_rate_pq(
  physeq,
  s = 0.1,
  nmin = 1000,
  smin = 3L,
  fmax = 0.02,
  cmin = 10L,
  rate = c("edgar2018", "fraction"),
  verbose = TRUE
)

Arguments

physeq: (phyloseq, required) A phyloseq object with integer counts.
s: (numeric, default 0.1) Low-count fraction of the OTU mean.
nmin: (numeric, default 1000) Minimum OTU total to be a candidate.
smin: (integer, default 3) Minimum number of low counts to be a candidate.
fmax: (numeric, default 0.02) Candidates with an estimated rate above this plausible maximum are rejected.
cmin: (integer, default 10) Minimum number of candidates required; below this the estimate is undetermined (NA).
rate: (character, default "edgar2018") Per-OTU rate formula, see Description.
verbose: (logical, default TRUE) Print a one-line summary.

Value

A single numeric: the median de novo cross-talk rate, or NA_real_ when undetermined. The number of candidate OTUs is attached as the "n_candidates" attribute.

Agreement with usearch

This reimplementation follows the paper. The usearch v11 binary diverges from its own description in undocumented ways (an absolute low-count cap of ~8 rather than s*N_i/n, the X_i/N_i rate, and an iterative candidate selection), so the returned value is of the same order of magnitude as usearch's de novo estimate but is not bit-exact. On integer-count data the per-OTU X_i/N_i reproduces usearch's reported rate when the same low-count set is used; the median over the whole table does not. See uncross2_pq() for the score-side comparison.

References

Edgar, R.C. (2018) UNCROSS2: identification of cross-talk in 16S rRNA OTU tables. bioRxiv 400762. doi:10.1101/400762

Author

Adrien Taudière

Examples

if (FALSE) { # \dontrun{
  estimate_xtalk_rate_pq(data_fungi)
} # }