Bounded Non-Contiguous Subsequence Mining
In gp3sequences: Transparent Analysis of Ordered Categorical Sequences

knitr::opts_chunk$set(collapse = TRUE, comment = "#>")
library(gp3sequences)

Why bounded subsequences?

A non-contiguous subsequence preserves order while allowing intervening states. The implementation requires explicit motif length, maximum gap, maximum span, and combination limits. These constraints keep enumeration auditable and avoid silently searching an unbounded combinatorial space.

Synthetic sequences

sequences <- data.frame(
  sequence_id = rep(paste0("s", 1:6), each = 5L),
  sequence_order = rep(1:5, times = 6L),
  state = c(
    "A", "B", "C", "D", "E",
    "A", "C", "B", "D", "E",
    "A", "B", "D", "C", "E",
    "E", "D", "C", "B", "A",
    "E", "C", "D", "B", "A",
    "E", "D", "B", "C", "A"
  ),
  group = rep(rep(c("forward", "reverse"), each = 3L), each = 5L),
  stringsAsFactors = FALSE
)

Enumerate occurrences

occurrences <- extract_sequence_subsequences(
  sequences,
  metadata_cols = "group",
  min_length = 2L,
  max_length = 3L,
  max_gap = 2L,
  max_span = 4L,
  repeated_state_policy = "preserve"
)
head(occurrences)
attributes(occurrences)[c("n_sequences", "settings")]

Sequence-level prevalence

subsequence_summary <- summarise_sequence_subsequences(occurrences)
head(subsequence_summary, 10L)

frequent <- filter_sequence_subsequences(
  subsequence_summary,
  min_sequences = 2L,
  min_prevalence = 0.25,
  top_n = 12L
)
frequent

Group comparison

comparison <- compare_sequence_subsequences(
  occurrences,
  group_col = "group",
  p_adjust = "holm"
)
head(comparison, 10L)

The comparison is based on sequence-level presence, not occurrence multiplicity. Adjusted p-values do not turn an observational grouping into a causal design.

plot_sequence_subsequences(frequent, metric = "sequence_prevalence")

Relation to specialist packages

The bounded enumerator is intentionally narrow. Large-scale frequent sequence mining, event-sequence constraint systems, and discriminating-subsequence algorithms remain appropriate uses of specialist packages through explicit adapters.