OBIKmers/obikmer
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

refactor: implement RoutableSuperKmer and update k-mer indexing pipeline

Browse Source 
Replace raw SuperkMer routing with a new RoutableSuperKimer type that embeds canonical sequences and precomputed minimizers, enabling direct partition routing via hash. Update the build pipeline to yield RoutableSuperKmers throughout (builder, scatterer), refactor FASTA/unitig export commands to use the new type and compressed outputs (.fasta.gz, .unitigs.fasta.zst), revise SuperKmer header to store n_kmers instead of seql (avoiding 256-byte wrap), and update documentation to reflect minimizer-based theory, two evidence-encoding strategies for unitig-MPHF indexing (global offset vs. ID+rank), and the new obipipeline library architecture with parallel workers, biased scheduling, and error handling.
This commit is contained in:
Eric Coissac
2026-04-29 22:52:42 +02:00
parent 4e26e3bd40
commit 27f5e88a7b
72 changed files with 10093 additions and 1626 deletions
Download Patch File Download Diff File Expand all files Collapse all files
doc/implementation/superkmer/index.html
+198 -12
View File
@@ -230,7 +230,7 @@
<li class="md-nav__item">
<a href="../../theory/kmers/" class="md-nav__link">
<a href="../../kmers/" class="md-nav__link">
@@ -313,6 +313,34 @@
<li class="md-nav__item">
<a href="../../theory/minimizer/" class="md-nav__link">
<span class="md-ellipsis">
Minimizer selection
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../theory/indexing/" class="md-nav__link">
@@ -488,6 +516,17 @@
</span>
</a>
</li>
<li class="md-nav__item">
<a href="#minimizer-sliding-window" class="md-nav__link">
<span class="md-ellipsis">
Minimizer sliding window
</span>
</a>
</li>
<li class="md-nav__item">
@@ -600,6 +639,34 @@
<li class="md-nav__item">
<a href="../obipipeline/" class="md-nav__link">
<span class="md-ellipsis">
obipipeline library
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../storage/" class="md-nav__link">
@@ -650,6 +717,34 @@
<li class="md-nav__item">
<a href="../unitig_evidence/" class="md-nav__link">
<span class="md-ellipsis">
Unitig evidence encoding
</span>
</a>
</li>
</ul>
</nav>
@@ -796,6 +891,17 @@
</span>
</a>
</li>
<li class="md-nav__item">
<a href="#minimizer-sliding-window" class="md-nav__link">
<span class="md-ellipsis">
Minimizer sliding window
</span>
</a>
</li>
<li class="md-nav__item">
@@ -828,7 +934,7 @@
<h1 id="superkmer-implementation">SuperKmer — implementation</h1>
<h2 id="memory-layout">Memory layout</h2>
<p>A super-kmer is stored as a <strong>32-bit header</strong> followed by a <strong>byte-aligned nucleotide sequence</strong> (2 bits/base, nucleotide 0 at the MSB of the first byte, max 256 nt):</p>
<p>A super-kmer is stored as a <strong>32-bit header</strong> followed by a <strong>byte-aligned nucleotide sequence</strong> (2 bits/base, nucleotide 0 at the MSB of the first byte):</p>
<table>
<thead>
<tr>
@@ -844,21 +950,44 @@
<td>Occurrence count (≤ 16 M)</td>
</tr>
<tr>
<td>SEQL</td>
<td>NKMERS</td>
<td>8</td>
<td>Sequence length in nucleotides (1–256)</td>
<td>Number of kmers (= seq_length − k + 1, range 1–255)</td>
</tr>
</tbody>
</table>
<p>Bit layout (MSB to LSB): <code>[31:8] COUNT [7:0] SEQL</code></p>
<p>SEQL is stored as a raw <code>u8</code>: values 1–255 represent lengths 1–255; <strong>0 represents 256</strong> (wrapping convention). The public accessor returns a <code>usize</code> and performs the conversion:</p>
<div class="highlight"><pre><span></span><code><span class="k">fn</span><span class="w"> </span><span class="nf">seql</span><span class="p">(</span><span class="o">&amp;</span><span class="bp">self</span><span class="p">)</span><span class="w"> </span><span class="p">-&gt;</span><span class="w"> </span><span class="kt">usize</span><span class="w"> </span><span class="p">{</span><span class="w"> </span><span class="k">if</span><span class="w"> </span><span class="n">s</span><span class="w"> </span><span class="o">==</span><span class="w"> </span><span class="mi">0</span><span class="w"> </span><span class="p">{</span><span class="w"> </span><span class="mi">256</span><span class="w"> </span><span class="p">}</span><span class="w"> </span><span class="k">else</span><span class="w"> </span><span class="p">{</span><span class="w"> </span><span class="n">s</span><span class="w"> </span><span class="k">as</span><span class="w"> </span><span class="kt">usize</span><span class="w"> </span><span class="p">}</span><span class="w"> </span><span class="p">}</span>
<p>Bit layout (MSB to LSB): <code>[31:8] COUNT [7:0] NKMERS</code></p>
<p>NKMERS is stored as a raw <code>u8</code> in <strong>kmer units</strong>, not nucleotides. The nucleotide length is recovered as <code>NKMERS + k − 1</code>. This avoids the awkward wrapping convention (<code>0 = 256</code>) that would be needed if nucleotide length were stored directly, and gains k−1 = 30 units of headroom:</p>
<table>
<thead>
<tr>
<th>unit</th>
<th>u8 covers</th>
<th>max nucleotides</th>
</tr>
</thead>
<tbody>
<tr>
<td>nucleotides</td>
<td>255 nt</td>
<td>225 kmers</td>
</tr>
<tr>
<td><strong>kmers</strong></td>
<td><strong>255 kmers</strong></td>
<td><strong>285 nt</strong></td>
</tr>
</tbody>
</table>
<p>The public accessors:</p>
<div class="highlight"><pre><span></span><code><span class="k">fn</span><span class="w"> </span><span class="nf">n_kmers</span><span class="p">(</span><span class="o">&amp;</span><span class="bp">self</span><span class="p">)</span><span class="w"> </span><span class="p">-&gt;</span><span class="w"> </span><span class="kt">usize</span><span class="w"> </span><span class="p">{</span><span class="w"> </span><span class="p">(</span><span class="bp">self</span><span class="p">.</span><span class="mi">0</span><span class="w"> </span><span class="o">&amp;</span><span class="w"> </span><span class="mh">0xFF</span><span class="p">)</span><span class="w"> </span><span class="k">as</span><span class="w"> </span><span class="kt">usize</span><span class="w"> </span><span class="p">}</span>
<span class="k">fn</span><span class="w"> </span><span class="nf">seql</span><span class="p">(</span><span class="o">&amp;</span><span class="bp">self</span><span class="p">)</span><span class="w"> </span><span class="p">-&gt;</span><span class="w"> </span><span class="kt">usize</span><span class="w"> </span><span class="p">{</span><span class="w"> </span><span class="bp">self</span><span class="p">.</span><span class="n">n_kmers</span><span class="p">()</span><span class="w"> </span><span class="o">+</span><span class="w"> </span><span class="n">K</span><span class="w"> </span><span class="o">-</span><span class="w"> </span><span class="mi">1</span><span class="w"> </span><span class="p">}</span>
<span class="k">fn</span><span class="w"> </span><span class="nf">count</span><span class="p">(</span><span class="o">&amp;</span><span class="bp">self</span><span class="p">)</span><span class="w"> </span><span class="p">-&gt;</span><span class="w"> </span><span class="kt">u32</span><span class="w"> </span><span class="p">{</span><span class="w"> </span><span class="bp">self</span><span class="p">.</span><span class="mi">0</span><span class="w"> </span><span class="o">&gt;&gt;</span><span class="w"> </span><span class="mi">8</span><span class="w"> </span><span class="p">}</span>
<span class="k">fn</span><span class="w"> </span><span class="nf">increment</span><span class="p">(</span><span class="o">&amp;</span><span class="k">mut</span><span class="w"> </span><span class="bp">self</span><span class="p">)</span><span class="w"> </span><span class="p">{</span><span class="w"> </span><span class="bp">self</span><span class="p">.</span><span class="mi">0</span><span class="w"> </span><span class="o">+=</span><span class="w"> </span><span class="mi">1</span><span class="w"> </span><span class="o">&lt;&lt;</span><span class="w"> </span><span class="mi">8</span><span class="p">;</span><span class="w"> </span><span class="p">}</span>
<span class="k">fn</span><span class="w"> </span><span class="nf">add</span><span class="p">(</span><span class="o">&amp;</span><span class="k">mut</span><span class="w"> </span><span class="bp">self</span><span class="p">,</span><span class="w"> </span><span class="n">n</span><span class="p">:</span><span class="w"> </span><span class="kt">u32</span><span class="p">)</span><span class="w"> </span><span class="p">{</span><span class="w"> </span><span class="bp">self</span><span class="p">.</span><span class="mi">0</span><span class="w"> </span><span class="o">+=</span><span class="w"> </span><span class="n">n</span><span class="w"> </span><span class="o">&lt;&lt;</span><span class="w"> </span><span class="mi">8</span><span class="p">;</span><span class="w"> </span><span class="p">}</span>
<span class="k">fn</span><span class="w"> </span><span class="nf">set_count</span><span class="p">(</span><span class="o">&amp;</span><span class="k">mut</span><span class="w"> </span><span class="bp">self</span><span class="p">,</span><span class="w"> </span><span class="n">n</span><span class="p">:</span><span class="w"> </span><span class="kt">u32</span><span class="p">)</span><span class="w"> </span><span class="p">{</span><span class="w"> </span><span class="bp">self</span><span class="p">.</span><span class="mi">0</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="p">(</span><span class="bp">self</span><span class="p">.</span><span class="mi">0</span><span class="w"> </span><span class="o">&amp;</span><span class="w"> </span><span class="mh">0xFF</span><span class="p">)</span><span class="w"> </span><span class="o">|</span><span class="w"> </span><span class="p">(</span><span class="n">n</span><span class="w"> </span><span class="o">&lt;&lt;</span><span class="w"> </span><span class="mi">8</span><span class="p">);</span><span class="w"> </span><span class="p">}</span>
</code></pre></div>
<p>The SEQL field is 8 bits, capping the stored sequence at 256 nt. Given the expected length of ~40 nt, this cap is almost never reached; when it is, the super-kmer is split at 256 nt with a k−1 overlap, preserving all kmers without duplication.</p>
<p>In practice, observed super-kmer lengths on metagenomic data (k=31) are below 55 nucleotides (≤ 25 kmers) — far from the 255-kmer cap. If a super-kmer ever exceeds 255 kmers, it is split with a k−1 nucleotide overlap, preserving all kmers without duplication (identical mechanism to partition-boundary splits).</p>
<p>The sequence is always stored in canonical form (lexicographic minimum of forward and reverse complement), with nucleotide 0 at the MSB of the first byte. The byte array can be hashed directly without any adjustment.</p>
<h2 id="ascii-encoding-and-decoding">ASCII encoding and decoding</h2>
<p>Two lookup tables handle ASCII ↔ 2-bit conversion:</p>
@@ -883,8 +1012,9 @@
<span class="p">}</span>
</code></pre></div>
<p><code>REVCOMP4</code> is 256 bytes (fits in L1 cache), computed at compile time. No endianness dependency — all operations are pure arithmetic on byte values.</p>
<p><strong>Step 2 — realignment.</strong> After step 1, <code>padding = n × 8 − SEQL × 2</code> spurious bits (complements of the original padding A's) appear at the start of the array. They are flushed left using <code>BitSlice&lt;u8, Msb0&gt;::rotate_left(padding)</code> from the <code>bitvec</code> crate, which is SIMD-accelerated. The trailing <code>padding</code> bits are then zeroed:</p>
<div class="highlight"><pre><span></span><code><span class="n">shift</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">n</span><span class="w"> </span><span class="o">*</span><span class="w"> </span><span class="mi">8</span><span class="w"> </span><span class="o">-</span><span class="w"> </span><span class="n">SEQL</span><span class="w"> </span><span class="o">*</span><span class="w"> </span><span class="mi">2</span><span class="w"> </span><span class="c1">// number of padding bits</span>
<p><strong>Step 2 — realignment.</strong> After step 1, <code>padding = n × 8 − seql × 2</code> spurious bits (complements of the original padding A's) appear at the start of the array. They are flushed left using <code>BitSlice&lt;u8, Msb0&gt;::rotate_left(padding)</code> from the <code>bitvec</code> crate, which is SIMD-accelerated. The trailing <code>padding</code> bits are then zeroed:</p>
<div class="highlight"><pre><span></span><code><span class="kd">let</span><span class="w"> </span><span class="n">seql</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="bp">self</span><span class="p">.</span><span class="n">n_kmers</span><span class="p">()</span><span class="w"> </span><span class="o">+</span><span class="w"> </span><span class="n">k</span><span class="w"> </span><span class="o">-</span><span class="w"> </span><span class="mi">1</span><span class="p">;</span>
<span class="n">shift</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">n</span><span class="w"> </span><span class="o">*</span><span class="w"> </span><span class="mi">8</span><span class="w"> </span><span class="o">-</span><span class="w"> </span><span class="n">seql</span><span class="w"> </span><span class="o">*</span><span class="w"> </span><span class="mi">2</span><span class="w"> </span><span class="c1">// number of padding bits</span>
<span class="n">bits</span><span class="p">.</span><span class="n">rotate_left</span><span class="p">(</span><span class="n">shift</span><span class="p">)</span>
<span class="n">bits</span><span class="p">[</span><span class="n">len</span><span class="w"> </span><span class="o">-</span><span class="w"> </span><span class="n">shift</span><span class="o">..</span><span class="p">].</span><span class="n">fill</span><span class="p">(</span><span class="kc">false</span><span class="p">)</span>
</code></pre></div>
@@ -900,6 +1030,61 @@
return seq -- palindrome: either orientation valid
</code></pre></div>
</div>
<h2 id="minimizer-sliding-window">Minimizer sliding window</h2>
<p>Super-kmers are built by <code>SuperKmerIter</code> (crate <code>obiskbuilder</code>), which maintains the current minimizer with a <strong>monotonic deque</strong> over a sliding window of W = k − m + 1 m-mer positions.</p>
<p>Each deque entry stores:</p>
<table>
<thead>
<tr>
<th>Field</th>
<th>Type</th>
<th>Purpose</th>
</tr>
</thead>
<tbody>
<tr>
<td><code>position</code></td>
<td>usize</td>
<td>0-based start of this m-mer in the segment</td>
</tr>
<tr>
<td><code>canonical</code></td>
<td>u64</td>
<td>right-aligned canonical m-mer value (lex-min of fwd and rc); used as partition key</td>
</tr>
<tr>
<td><code>hash</code></td>
<td>u64</td>
<td><span class="arithmatex">\(H(\text{canonical})\)</span> — ordering key for random minimizer selection</td>
</tr>
</tbody>
</table>
<p>The hash <span class="arithmatex">\(H\)</span> is the seeded splitmix64 finalizer (see <a href="../../theory/minimizer/">Minimizer selection</a>):</p>
<div class="highlight"><pre><span></span><code><span class="k">fn</span><span class="w"> </span><span class="nf">hash_mmer</span><span class="p">(</span><span class="n">canonical</span><span class="p">:</span><span class="w"> </span><span class="kt">u64</span><span class="p">)</span><span class="w"> </span><span class="p">-&gt;</span><span class="w"> </span><span class="kt">u64</span><span class="w"> </span><span class="p">{</span>
<span class="w"> </span><span class="kd">let</span><span class="w"> </span><span class="n">x</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">canonical</span><span class="w"> </span><span class="o">^</span><span class="w"> </span><span class="mh">0x9e3779b97f4a7c15</span><span class="p">;</span><span class="w"> </span><span class="c1">// seed: eliminates fixed point at 0</span>
<span class="w"> </span><span class="kd">let</span><span class="w"> </span><span class="n">x</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">x</span><span class="w"> </span><span class="o">^</span><span class="w"> </span><span class="p">(</span><span class="n">x</span><span class="w"> </span><span class="o">&gt;&gt;</span><span class="w"> </span><span class="mi">30</span><span class="p">);</span>
<span class="w"> </span><span class="kd">let</span><span class="w"> </span><span class="n">x</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">x</span><span class="p">.</span><span class="n">wrapping_mul</span><span class="p">(</span><span class="mh">0xbf58476d1ce4e5b9</span><span class="p">);</span>
<span class="w"> </span><span class="kd">let</span><span class="w"> </span><span class="n">x</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">x</span><span class="w"> </span><span class="o">^</span><span class="w"> </span><span class="p">(</span><span class="n">x</span><span class="w"> </span><span class="o">&gt;&gt;</span><span class="w"> </span><span class="mi">27</span><span class="p">);</span>
<span class="w"> </span><span class="kd">let</span><span class="w"> </span><span class="n">x</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">x</span><span class="p">.</span><span class="n">wrapping_mul</span><span class="p">(</span><span class="mh">0x94d049bb133111eb</span><span class="p">);</span>
<span class="w"> </span><span class="n">x</span><span class="w"> </span><span class="o">^</span><span class="w"> </span><span class="p">(</span><span class="n">x</span><span class="w"> </span><span class="o">&gt;&gt;</span><span class="w"> </span><span class="mi">31</span><span class="p">)</span>
<span class="p">}</span>
</code></pre></div>
<p>On each new nucleotide, once the window is full, the deque is updated:</p>
<div class="admonition abstract">
<p class="admonition-title">Algorithm — minimizer deque update</p>
<div class="highlight"><pre><span></span><code>procedure UpdateMinimizer(deque, position, canonical, hash, k, received):
-- pop dominated entries from the back
while deque.back.hash ≥ hash:
deque.pop_back()
deque.push_back({position, canonical, hash})
-- evict expired entries from the front
while deque.front.position + k &lt; received:
deque.pop_front()
</code></pre></div>
</div>
<p>The front of the deque is always the current minimizer. Because the deque is maintained in strictly increasing hash order, each entry is popped at most once — O(1) amortized per nucleotide.</p>
<p>A super-kmer boundary is emitted when the minimizer changes: <code>deque.front.hash ≠ prev_hash</code>. The <code>canonical</code> field of the front entry is <strong>not</strong> used for boundary detection — that uses the hash alone. The canonical value is stored so that the partition key <span class="arithmatex">\(H(\text{canonical})\)</span> can be recomputed independently at routing time from the stored <code>minimizer_pos</code>, without inheriting the minimum-order-statistic bias (see <a href="../../theory/minimizer/#partition-key-independence">Minimizer selection — partition key independence</a>).</p>
<h2 id="kmer-extraction">Kmer extraction</h2>
<p>A k-mer is extracted from a super-kmer with <code>SuperKmer::kmer(i, k)</code>, which returns a <code>Kmer</code> — a left-aligned <code>u64</code> newtype (see <a href="../kmer/">Kmer implementation</a>):</p>
<div class="highlight"><pre><span></span><code><span class="k">pub</span><span class="w"> </span><span class="k">fn</span><span class="w"> </span><span class="nf">kmer</span><span class="p">(</span><span class="o">&amp;</span><span class="bp">self</span><span class="p">,</span><span class="w"> </span><span class="n">i</span><span class="p">:</span><span class="w"> </span><span class="kt">usize</span><span class="p">,</span><span class="w"> </span><span class="n">k</span><span class="p">:</span><span class="w"> </span><span class="kt">usize</span><span class="p">)</span><span class="w"> </span><span class="p">-&gt;</span><span class="w"> </span><span class="nb">Result</span><span class="o">&lt;</span><span class="n">Kmer</span><span class="p">,</span><span class="w"> </span><span class="n">KmerError</span><span class="o">&gt;</span>
@@ -909,8 +1094,9 @@
<div class="admonition abstract">
<p class="admonition-title">Algorithm — Super-kmer reverse complement</p>
<div class="highlight"><pre><span></span><code>procedure SuperKmerRevcomp(seq, SEQL):
n ← ⌈SEQL / 4⌉ -- number of bytes
shift ← n × 8 − SEQL × 2 -- padding bits to flush
seql ← NKMERS + k − 1 -- nucleotide length
n ← ⌈seql / 4⌉ -- number of bytes
shift ← n × 8 − seql × 2 -- padding bits to flush
-- step 1: swap bytes outside-in, applying REVCOMP4 to each (256-byte L1 table)
lo ← 0 ; hi ← n − 1