countmut 0.3.2__tar.gz → 0.3.3__tar.gz

{countmut-0.3.2/countmut.egg-info → countmut-0.3.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: countmut
-Version: 0.3.2
+Version: 0.3.3
 Summary: Unified ultra-fast strand-aware mutation counter (C backend + samtools-style -e/-p filters)
 Author-email: Ye Chang <yech1990@gmail.com>
 License-Expression: MIT

{countmut-0.3.2 → countmut-0.3.3}/backend/countmut_core.c

@@ -96,6 +96,22 @@ static int base_to_index(char c) {
     }
 }
 
+/* Reference-frame base index for a read at a CIGAR query offset `qpos`.
+ *
+ * `qpos` is the query offset walked left->right along the CIGAR (reference
+ * order).  bwa / minibwa write the SEQ of a reverse-strand read already in
+ * reference-forward orientation (the stored SEQ is the reverse complement of
+ * the original read, matching the reference), so SEQ[qpos] is directly the
+ * reference-frame base -- no reverse-complement or index reversal is needed.
+ *
+ * Returns 0..3 for A/C/G/T, or 4 for any other symbol.
+ */
+static int ref_frame_base(const bam1_t *b, uint32_t qpos) {
+    uint32_t qlen = b->core.l_qseq;
+    if (qpos >= qlen) return 4;
+    return nt16_index(bam_seqi(bam_get_seq(b), qpos));
+}
+
 /* Reverse-complement a single uppercase motif base (A<->T, C<->G, else N). */
 static char rc_nt(char c) {
     switch (c) {
@@ -567,13 +583,10 @@ static void count_interval(worker_t *w, const cm_config *cfg, bam_hdr_t *hdr, FI
             const bam_pileup1_t *p = &plp[0][i];
             const bam1_t *b = p->b;
             int s = bio_strand(b);
-            uint8_t nt = bam_seqi(bam_get_seq(b), p->qpos);
-            int base_i = nt16_index(nt);
-            /* Reverse-oriented reads: stored SEQ is 5'->3' (SAM spec) but qpos
-             * walks CIGAR order (left->right); complement the base into the
-             * reference frame.  Key on bam_is_rev (read orientation), not the
-             * biological strand s (which differs for R2 in paired data). */
-            if (bam_is_rev(b) && base_i < 4) base_i = 3 - base_i;
+            /* Reference-frame base: for a reverse read the stored SEQ is the
+             * reverse complement, so the base lives at SEQ index qlen-1-qpos
+             * and must be complemented (see ref_frame_base). */
+            int base_i = ref_frame_base(b, (uint32_t)p->qpos);
             /* router-assigned category slot (0..CM_CAT_MAX-1); g_a set in
              * the selection loop for every kept candidate. */
             int cat = w->g_a[i];
@@ -763,15 +776,12 @@ static rw_w *rw_add_base(worker_t *w, const cm_config *cfg, bam_hdr_t *hdr, int
                               (ref_pos >= 0 && ref_pos < w->chr_len) ? w->chr_seq[ref_pos] : 'N');
         if (rslot < 0) return wins;
     }
-    uint8_t nt = bam_seqi(bam_get_seq(b), qpos);
-    int base_i = nt16_index(nt);
-    /* Reverse-oriented reads: stored SEQ is 5'->3' (SAM spec) but qpos walks
-     * CIGAR order (left->right); complement the base into the reference frame
-     * (parity with countmut 0.0.x + pysam pairs).  This must key on the read's
-     * own orientation (bam_is_rev), NOT on the biological strand s: for paired
-     * reads an R2 mapped forward has bio_strand=1 yet is NOT reverse-oriented,
-     * so complementing on s would wrongly flip its bases (A<->T). */
-    if (bam_is_rev(b) && base_i < 4) base_i = 3 - base_i;
+    /* Reference-frame base: for a reverse read the stored SEQ is the reverse
+     * complement, so the base lives at SEQ index qlen-1-qpos and must be
+     * complemented (see ref_frame_base).  Keying on the read's own orientation
+     * (bam_is_rev) is correct; the biological strand s differs for R2 in
+     * paired data. */
+    int base_i = ref_frame_base(b, qpos);
     int qual = (int)bam_get_qual(b)[qpos];
     if (direct) {
         int cat = rslot;
@@ -1042,11 +1052,7 @@ static void count_interval_readwalk(worker_t *w, const cm_config *cfg, bam_hdr_t
                             if (ref_pos < beg || ref_pos >= end) continue;
                             uint32_t qpos = qcur + (uint32_t)k;
                             if (qpos >= qlen) break;
-                            uint8_t nt = bam_seqi(bam_get_seq(b), qpos);
-                            int base_i = nt16_index(nt);
-                            /* Complement on read orientation (bam_is_rev), not
-                             * the biological strand s (R2 differs). */
-                            if (bam_is_rev(b) && base_i < 4) base_i = 3 - base_i;
+                            int base_i = ref_frame_base(b, qpos);
                             int cat = 0;
                             if (w->expr && cm_expr_has_read(w->expr) && !cm_expr_read_constant(w->expr)) {
                                 cat = cm_expr_route(w->expr, b, hdr->target_name[tid],
@@ -1114,11 +1120,7 @@ static void count_interval_readwalk(worker_t *w, const cm_config *cfg, bam_hdr_t
                                     uint32_t qpos = qcur + (uint32_t)k;
                                     if (qpos >= qlen) break;
                                     int in_ovl = ((int)qpos >= olo && (int)qpos < ohi);
-                                    uint8_t nt = bam_seqi(bam_get_seq(b), qpos);
-                                    int base_i = nt16_index(nt);
-                                    /* Complement on read orientation (bam_is_rev),
-                                     * not the biological strand s (R2 differs). */
-                                    if (bam_is_rev(b) && base_i < 4) base_i = 3 - base_i;
+                                    int base_i = ref_frame_base(b, qpos);
                                     int cat = 0;
                                     if (w->expr && cm_expr_has_read(w->expr) && !cm_expr_read_constant(w->expr)) {
                                         cat = cm_expr_route(w->expr, b, hdr->target_name[tid],

{countmut-0.3.2 → countmut-0.3.3/countmut.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: countmut
-Version: 0.3.2
+Version: 0.3.3
 Summary: Unified ultra-fast strand-aware mutation counter (C backend + samtools-style -e/-p filters)
 Author-email: Ye Chang <yech1990@gmail.com>
 License-Expression: MIT

{countmut-0.3.2 → countmut-0.3.3}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "countmut"
-version = "0.3.2"
+version = "0.3.3"
 description = "Unified ultra-fast strand-aware mutation counter (C backend + samtools-style -e/-p filters)"
 readme = "README.md"
 license = "MIT"

{countmut-0.3.2 → countmut-0.3.3}/tests/test_correctness.py

@@ -92,37 +92,33 @@ def indel_data(tmp_path_factory):
 
 
 # ---------------------------------------------------------------------------
-# BUG: base complement keyed on biological strand (s) instead of read
-# orientation (bam_is_rev).  For paired data an R2 read mapped forward has
-# bio_strand=1 yet is NOT reverse-oriented, so its bases were wrongly
-# reverse-complemented (A<->T), turning a pure-reference site into a 50/50
-# A/T mix.  Regression: R2-forward reads must count their stored bases as-is.
+# BUG: base complement applied to reverse-strand reads.  bwa / minibwa write
+# the SEQ of a reverse-strand read already in reference-forward orientation
+# (the stored SEQ is the reverse complement of the original read, matching the
+# reference), so the reference-frame base is simply SEQ[qpos].  Complementing
+# or index-reversing it (as an earlier fix did) wrongly flipped A<->T for every
+# reverse read, turning a pure-reference site into a ~50/50 A/T mix.
+# Regression: reverse-strand reads must be counted as-is.
 # ---------------------------------------------------------------------------
-def test_r2_forward_not_complemented(tmp_path):
-    root = str(tmp_path / "r2f")
+def test_reverse_read_not_complemented(tmp_path):
+    root = str(tmp_path / "rev")
     os.makedirs(root, exist_ok=True)
-    # reference base at 0-based 0 is 'A'.  An R2 read mapped FORWARD stores the
-    # base as-is (A), so a pure-A site must stay A, not flip to T.
+    # reference base at 0-based 0 is 'A'.  A reverse-strand read (flag 16) whose
+    # stored SEQ is reference-forward (as bwa/minibwa write it) carries 'A' at
+    # 0-based 0; it must stay A, not flip to T.
     bam, fa = _write_bam(
         root,
         [
-            # R1 forward (flag 99) and R2 forward (flag 35) covering 0-based 0.
-            lambda h: _mk(h, "frag1", 99, 0, "ACGTACGTAC"),
-            lambda h: _mk(h, "frag1", 35, 0, "ACGTACGTAC"),
-            # R1 reverse (flag 83) and R2 reverse (flag 147): stored SEQ is the
-            # reverse complement, so they must be complemented back to A.
-            lambda h: _mk(h, "frag2", 83, 0, "ACGTACGTAC"),
-            lambda h: _mk(h, "frag2", 147, 0, "ACGTACGTAC"),
+            lambda h: _mk(h, "fwd", 0, 0, "ACGTACGTAC"),
+            lambda h: _mk(h, "rev", 16, 0, "ACGTACGTAC"),
         ],
     )
-    _h, rows = run_c(bam, fa, engine="read-walk", region="chr1:1-5")
-    # rows: [chrom, pos, ref, depth, a, c, g, t, n]  (strandless default)
+    _h, rows = run_c(bam, fa, engine="read-walk", region="chr1:1-5",
+                     extra=["--strandless"])
     row = [r for r in rows if r[1] == 1][0]
     a, c, g, t = row[4], row[5], row[6], row[7]
-    # All four reads carry reference 'A' at 0-based 0 -> A must dominate.
-    assert a > 0 and a >= c + g + t, (row)
-    # The bug produced ~50/50 A/T; assert T is a small minority.
-    assert t <= a * 0.1, (row)
+    # Both reads carry reference 'A' at 0-based 0 -> A must dominate.
+    assert a == 2 and t == 0, (row)
 
 
 # ---------------------------------------------------------------------------
@@ -421,7 +417,6 @@ def _write_reference_read_bam(tmp_path, chrom, length):
     return bam, fa
 
 
-@pytest.mark.xfail(reason="full htslib SAM transcode produces non-BGZF temp BAM; CRAM/BAM paths are covered")
 def test_sam_input_matches_bam(motif_data, tmp_path):
     """SAM (plain, and gzipped) input must produce byte-identical output to the
     equivalent BAM (it is auto-transcoded to a temp BAM + index)."""

{countmut-0.3.2 → countmut-0.3.3}/tests/test_router.py

@@ -80,13 +80,15 @@ def data(tmp_path_factory):
         mk("p_g", 0, 0, "GGGGGAATT", Q40, ok),
         # p_x: NS > 1 -> dropped entirely
         mk("p_x", 0, 0, "GGGGAAATT", Q40, [("NS", 2), ("Yf", 1), ("Zf", 0)]),
-        # minus reads, 9M at 0-based 4..12; stored T -> reference-frame A.
-        # qpos = pos - 4, so group 1 (qpos 2..6) = 0-based 6..10.
-        mk("m_hi", 16, 4, "TTTTTTTTT", Q40, ok),
+        # minus reads, 9M at 0-based 4..12.  bwa/minibwa store the SEQ of a
+        # reverse-strand read in reference-forward orientation, so the stored
+        # base equals the reference base (no complement needed).  reference
+        # [4:13] = AAAACCCC + 1bp past the contig end.
+        mk("m_hi", 16, 4, "AAAACCCCC", Q40, ok),
         # m_lo: no Yf -> group 0 at every base
-        mk("m_lo", 16, 4, "TTTTTTTTT", Q40, [("NS", 0), ("Yf", 0), ("Zf", 0)]),
+        mk("m_lo", 16, 4, "AAAACCCCC", Q40, [("NS", 0), ("Yf", 0), ("Zf", 0)]),
         # m_x: NS > 1 -> dropped entirely
-        mk("m_x", 16, 4, "TTTTTTTTT", Q40, [("NS", 2), ("Yf", 1), ("Zf", 0)]),
+        mk("m_x", 16, 4, "AAAACCCCC", Q40, [("NS", 2), ("Yf", 1), ("Zf", 0)]),
     ]
     reads.sort(key=lambda r: r.reference_start)
     with pysam.AlignmentFile(bam, "wb", header=header) as out: