512 lines
15 KiB
Rust
512 lines
15 KiB
Rust
//! K-mer encoding and HNSW vector indexing for DNA sequences
|
|
//!
|
|
//! This module provides efficient k-mer based vector encoding for DNA sequences
|
|
//! with HNSW indexing for fast similarity search. Implements both k-mer frequency
|
|
//! vectors and MinHash sketching (Mash/sourmash algorithm).
|
|
|
|
use ruvector_core::{
|
|
types::{DbOptions, DistanceMetric, HnswConfig, QuantizationConfig, SearchQuery},
|
|
VectorDB, VectorEntry,
|
|
};
|
|
use std::collections::HashMap;
|
|
use thiserror::Error;
|
|
|
|
#[derive(Error, Debug)]
|
|
pub enum KmerError {
|
|
#[error("Invalid k-mer length: {0}")]
|
|
InvalidKmerLength(usize),
|
|
#[error("Invalid DNA sequence: {0}")]
|
|
InvalidSequence(String),
|
|
#[error("Database error: {0}")]
|
|
DatabaseError(#[from] ruvector_core::RuvectorError),
|
|
#[error("Empty sequence")]
|
|
EmptySequence,
|
|
}
|
|
|
|
type Result<T> = std::result::Result<T, KmerError>;
|
|
|
|
/// Nucleotide to 2-bit encoding: A=0, C=1, G=2, T=3
|
|
#[inline]
|
|
fn nucleotide_to_bits(nuc: u8) -> Option<u8> {
|
|
match nuc.to_ascii_uppercase() {
|
|
b'A' => Some(0),
|
|
b'C' => Some(1),
|
|
b'G' => Some(2),
|
|
b'T' | b'U' => Some(3),
|
|
_ => None,
|
|
}
|
|
}
|
|
|
|
/// Returns the reverse complement of a DNA sequence
|
|
fn reverse_complement(seq: &[u8]) -> Vec<u8> {
|
|
seq.iter()
|
|
.rev()
|
|
.map(|&nuc| match nuc.to_ascii_uppercase() {
|
|
b'A' => b'T',
|
|
b'T' | b'U' => b'A',
|
|
b'C' => b'G',
|
|
b'G' => b'C',
|
|
n => n,
|
|
})
|
|
.collect()
|
|
}
|
|
|
|
/// Returns the canonical k-mer (lexicographically smaller of k-mer and its reverse complement)
|
|
pub fn canonical_kmer(kmer: &[u8]) -> Vec<u8> {
|
|
let rc = reverse_complement(kmer);
|
|
if kmer <= rc.as_slice() {
|
|
kmer.to_vec()
|
|
} else {
|
|
rc
|
|
}
|
|
}
|
|
|
|
/// K-mer encoder that converts DNA sequences into frequency vectors
|
|
pub struct KmerEncoder {
|
|
k: usize,
|
|
dimensions: usize,
|
|
}
|
|
|
|
impl KmerEncoder {
|
|
/// Create a new k-mer encoder for k-mers of length k
|
|
///
|
|
/// # Arguments
|
|
/// * `k` - Length of k-mers (typical values: 21, 31)
|
|
///
|
|
/// Uses feature hashing to limit dimensionality for large k
|
|
pub fn new(k: usize) -> Result<Self> {
|
|
if k == 0 || k > 32 {
|
|
return Err(KmerError::InvalidKmerLength(k));
|
|
}
|
|
|
|
// Calculate dimensions: min(4^k, 1024) using feature hashing
|
|
let max_kmers = 4_usize.saturating_pow(k as u32);
|
|
let dimensions = max_kmers.min(1024);
|
|
|
|
Ok(Self { k, dimensions })
|
|
}
|
|
|
|
/// Get the number of dimensions in the encoded vector
|
|
pub fn dimensions(&self) -> usize {
|
|
self.dimensions
|
|
}
|
|
|
|
/// Encode a DNA sequence into a k-mer frequency vector
|
|
///
|
|
/// Uses canonical k-mer hashing (min of forward/reverse-complement hash)
|
|
/// to count strand-agnostic k-mers, then normalizes to unit vector.
|
|
pub fn encode_sequence(&self, seq: &[u8]) -> Result<Vec<f32>> {
|
|
if seq.len() < self.k {
|
|
return Err(KmerError::EmptySequence);
|
|
}
|
|
|
|
let mut counts = vec![0u32; self.dimensions];
|
|
let mut total = 0u32;
|
|
|
|
// Extract all k-mers using a sliding window
|
|
// Avoid Vec allocation by hashing both strands and taking min
|
|
for window in seq.windows(self.k) {
|
|
let fwd_hash = Self::fnv1a_hash(window);
|
|
let rc_hash = Self::fnv1a_hash_rc(window);
|
|
let canonical_hash = fwd_hash.min(rc_hash);
|
|
let index = canonical_hash % self.dimensions;
|
|
|
|
counts[index] = counts[index].saturating_add(1);
|
|
total = total.saturating_add(1);
|
|
}
|
|
|
|
// Normalize to frequency vector and then to unit vector
|
|
let inv_total = 1.0 / total as f32;
|
|
let mut vector: Vec<f32> = counts
|
|
.iter()
|
|
.map(|&count| count as f32 * inv_total)
|
|
.collect();
|
|
|
|
// L2 normalization
|
|
let norm: f32 = vector.iter().map(|x| x * x).sum::<f32>().sqrt();
|
|
if norm > 0.0 {
|
|
let inv_norm = 1.0 / norm;
|
|
vector.iter_mut().for_each(|x| *x *= inv_norm);
|
|
}
|
|
|
|
Ok(vector)
|
|
}
|
|
|
|
/// FNV-1a hash of a byte slice
|
|
#[inline]
|
|
fn fnv1a_hash(data: &[u8]) -> usize {
|
|
const FNV_OFFSET: u64 = 14695981039346656037;
|
|
const FNV_PRIME: u64 = 1099511628211;
|
|
let mut hash = FNV_OFFSET;
|
|
for &byte in data {
|
|
hash ^= byte as u64;
|
|
hash = hash.wrapping_mul(FNV_PRIME);
|
|
}
|
|
hash as usize
|
|
}
|
|
|
|
/// FNV-1a hash of reverse complement (avoids Vec allocation)
|
|
#[inline]
|
|
fn fnv1a_hash_rc(data: &[u8]) -> usize {
|
|
const FNV_OFFSET: u64 = 14695981039346656037;
|
|
const FNV_PRIME: u64 = 1099511628211;
|
|
let mut hash = FNV_OFFSET;
|
|
for &byte in data.iter().rev() {
|
|
let comp = match byte.to_ascii_uppercase() {
|
|
b'A' => b'T',
|
|
b'T' | b'U' => b'A',
|
|
b'C' => b'G',
|
|
b'G' => b'C',
|
|
n => n,
|
|
};
|
|
hash ^= comp as u64;
|
|
hash = hash.wrapping_mul(FNV_PRIME);
|
|
}
|
|
hash as usize
|
|
}
|
|
|
|
/// Hash a k-mer to an index using FNV-1a hash
|
|
fn hash_kmer(&self, kmer: &[u8]) -> usize {
|
|
Self::fnv1a_hash(kmer)
|
|
}
|
|
}
|
|
|
|
/// MinHash sketch for fast sequence similarity (Mash/sourmash algorithm)
|
|
pub struct MinHashSketch {
|
|
num_hashes: usize,
|
|
hashes: Vec<u64>,
|
|
}
|
|
|
|
impl MinHashSketch {
|
|
/// Create a new MinHash sketch with the given number of hashes
|
|
///
|
|
/// # Arguments
|
|
/// * `num_hashes` - Number of hash values to keep (typically 1000)
|
|
pub fn new(num_hashes: usize) -> Self {
|
|
Self {
|
|
num_hashes,
|
|
hashes: Vec::new(),
|
|
}
|
|
}
|
|
|
|
/// Compute MinHash signature for a DNA sequence
|
|
pub fn sketch(&mut self, seq: &[u8], k: usize) -> Result<&[u64]> {
|
|
if seq.len() < k {
|
|
return Err(KmerError::EmptySequence);
|
|
}
|
|
|
|
let mut all_hashes = Vec::with_capacity(seq.len() - k + 1);
|
|
|
|
// Hash all k-mers using dual-hash (no Vec allocation per k-mer)
|
|
for window in seq.windows(k) {
|
|
let fwd = Self::hash_kmer_64_slice(window);
|
|
let rc = Self::hash_kmer_64_rc(window);
|
|
all_hashes.push(fwd.min(rc));
|
|
}
|
|
|
|
// Sort and keep the smallest num_hashes values
|
|
all_hashes.sort_unstable();
|
|
all_hashes.truncate(self.num_hashes);
|
|
self.hashes = all_hashes;
|
|
|
|
Ok(&self.hashes)
|
|
}
|
|
|
|
/// Compute Jaccard distance between two MinHash sketches
|
|
pub fn jaccard_distance(&self, other: &MinHashSketch) -> f32 {
|
|
if self.hashes.is_empty() || other.hashes.is_empty() {
|
|
return 1.0;
|
|
}
|
|
|
|
let mut intersection = 0;
|
|
let mut i = 0;
|
|
let mut j = 0;
|
|
|
|
// Count intersection using sorted arrays
|
|
while i < self.hashes.len() && j < other.hashes.len() {
|
|
if self.hashes[i] == other.hashes[j] {
|
|
intersection += 1;
|
|
i += 1;
|
|
j += 1;
|
|
} else if self.hashes[i] < other.hashes[j] {
|
|
i += 1;
|
|
} else {
|
|
j += 1;
|
|
}
|
|
}
|
|
|
|
let union = self.hashes.len() + other.hashes.len() - intersection;
|
|
if union == 0 {
|
|
return 0.0;
|
|
}
|
|
|
|
let jaccard_similarity = intersection as f32 / union as f32;
|
|
1.0 - jaccard_similarity
|
|
}
|
|
|
|
/// Hash a k-mer using MurmurHash3-like algorithm (forward strand)
|
|
#[inline]
|
|
fn hash_kmer_64_slice(kmer: &[u8]) -> u64 {
|
|
const C1: u64 = 0x87c37b91114253d5;
|
|
const C2: u64 = 0x4cf5ad432745937f;
|
|
let mut h = 0u64;
|
|
for &byte in kmer {
|
|
let mut k = byte as u64;
|
|
k = k.wrapping_mul(C1);
|
|
k = k.rotate_left(31);
|
|
k = k.wrapping_mul(C2);
|
|
h ^= k;
|
|
h = h.rotate_left(27);
|
|
h = h.wrapping_mul(5).wrapping_add(0x52dce729);
|
|
}
|
|
h ^ kmer.len() as u64
|
|
}
|
|
|
|
/// Hash reverse complement of a k-mer (no Vec allocation)
|
|
#[inline]
|
|
fn hash_kmer_64_rc(kmer: &[u8]) -> u64 {
|
|
const C1: u64 = 0x87c37b91114253d5;
|
|
const C2: u64 = 0x4cf5ad432745937f;
|
|
let mut h = 0u64;
|
|
for &byte in kmer.iter().rev() {
|
|
let comp = match byte.to_ascii_uppercase() {
|
|
b'A' => b'T',
|
|
b'T' | b'U' => b'A',
|
|
b'C' => b'G',
|
|
b'G' => b'C',
|
|
n => n,
|
|
};
|
|
let mut k = comp as u64;
|
|
k = k.wrapping_mul(C1);
|
|
k = k.rotate_left(31);
|
|
k = k.wrapping_mul(C2);
|
|
h ^= k;
|
|
h = h.rotate_left(27);
|
|
h = h.wrapping_mul(5).wrapping_add(0x52dce729);
|
|
}
|
|
h ^ kmer.len() as u64
|
|
}
|
|
|
|
/// Get the hashes
|
|
pub fn hashes(&self) -> &[u64] {
|
|
&self.hashes
|
|
}
|
|
}
|
|
|
|
/// Search result for k-mer index queries
|
|
#[derive(Debug, Clone)]
|
|
pub struct KmerSearchResult {
|
|
pub id: String,
|
|
pub score: f32,
|
|
pub distance: f32,
|
|
}
|
|
|
|
/// K-mer index wrapping VectorDB for sequence similarity search
|
|
pub struct KmerIndex {
|
|
db: VectorDB,
|
|
encoder: KmerEncoder,
|
|
k: usize,
|
|
}
|
|
|
|
impl KmerIndex {
|
|
/// Create a new k-mer index
|
|
///
|
|
/// # Arguments
|
|
/// * `k` - K-mer length
|
|
/// * `dimensions` - Vector dimensions (should match encoder dimensions)
|
|
pub fn new(k: usize, dimensions: usize) -> Result<Self> {
|
|
let encoder = KmerEncoder::new(k)?;
|
|
|
|
// Verify dimensions match
|
|
if encoder.dimensions() != dimensions {
|
|
return Err(KmerError::InvalidKmerLength(k));
|
|
}
|
|
|
|
let options = DbOptions {
|
|
dimensions,
|
|
distance_metric: DistanceMetric::Cosine,
|
|
storage_path: format!("./kmer_index_k{}.db", k),
|
|
hnsw_config: Some(HnswConfig {
|
|
m: 32,
|
|
ef_construction: 200,
|
|
ef_search: 100,
|
|
max_elements: 1_000_000,
|
|
}),
|
|
quantization: Some(QuantizationConfig::Scalar),
|
|
};
|
|
|
|
let db = VectorDB::new(options)?;
|
|
|
|
Ok(Self { db, encoder, k })
|
|
}
|
|
|
|
/// Index a single DNA sequence
|
|
pub fn index_sequence(&self, id: &str, sequence: &[u8]) -> Result<()> {
|
|
let vector = self.encoder.encode_sequence(sequence)?;
|
|
|
|
let entry = VectorEntry {
|
|
id: Some(id.to_string()),
|
|
vector,
|
|
metadata: Some({
|
|
let mut meta = HashMap::new();
|
|
meta.insert("length".to_string(), serde_json::json!(sequence.len()));
|
|
meta.insert("k".to_string(), serde_json::json!(self.k));
|
|
meta
|
|
}),
|
|
};
|
|
|
|
self.db.insert(entry)?;
|
|
Ok(())
|
|
}
|
|
|
|
/// Index multiple sequences in a batch
|
|
pub fn index_batch(&self, sequences: Vec<(&str, &[u8])>) -> Result<()> {
|
|
let entries: Result<Vec<VectorEntry>> = sequences
|
|
.into_iter()
|
|
.map(|(id, seq)| {
|
|
let vector = self.encoder.encode_sequence(seq)?;
|
|
Ok(VectorEntry {
|
|
id: Some(id.to_string()),
|
|
vector,
|
|
metadata: Some({
|
|
let mut meta = HashMap::new();
|
|
meta.insert("length".to_string(), serde_json::json!(seq.len()));
|
|
meta.insert("k".to_string(), serde_json::json!(self.k));
|
|
meta
|
|
}),
|
|
})
|
|
})
|
|
.collect();
|
|
|
|
self.db.insert_batch(entries?)?;
|
|
Ok(())
|
|
}
|
|
|
|
/// Search for similar sequences
|
|
pub fn search_similar(&self, query: &[u8], top_k: usize) -> Result<Vec<KmerSearchResult>> {
|
|
let query_vector = self.encoder.encode_sequence(query)?;
|
|
|
|
let search_query = SearchQuery {
|
|
vector: query_vector,
|
|
k: top_k,
|
|
filter: None,
|
|
ef_search: None,
|
|
};
|
|
|
|
let results = self.db.search(search_query)?;
|
|
|
|
Ok(results
|
|
.into_iter()
|
|
.map(|r| KmerSearchResult {
|
|
id: r.id,
|
|
score: r.score,
|
|
distance: r.score,
|
|
})
|
|
.collect())
|
|
}
|
|
|
|
/// Search for sequences with similarity above a threshold
|
|
pub fn search_with_threshold(
|
|
&self,
|
|
query: &[u8],
|
|
threshold: f32,
|
|
) -> Result<Vec<KmerSearchResult>> {
|
|
// Search with a larger k to ensure we get all candidates
|
|
let results = self.search_similar(query, 100)?;
|
|
|
|
Ok(results
|
|
.into_iter()
|
|
.filter(|r| r.distance <= threshold)
|
|
.collect())
|
|
}
|
|
|
|
/// Get the k-mer length
|
|
pub fn k(&self) -> usize {
|
|
self.k
|
|
}
|
|
}
|
|
|
|
#[cfg(test)]
|
|
mod tests {
|
|
use super::*;
|
|
|
|
#[test]
|
|
fn test_nucleotide_encoding() {
|
|
assert_eq!(nucleotide_to_bits(b'A'), Some(0));
|
|
assert_eq!(nucleotide_to_bits(b'C'), Some(1));
|
|
assert_eq!(nucleotide_to_bits(b'G'), Some(2));
|
|
assert_eq!(nucleotide_to_bits(b'T'), Some(3));
|
|
assert_eq!(nucleotide_to_bits(b'a'), Some(0));
|
|
assert_eq!(nucleotide_to_bits(b'N'), None);
|
|
}
|
|
|
|
#[test]
|
|
fn test_reverse_complement() {
|
|
let seq = b"ATCG";
|
|
let rc = reverse_complement(seq);
|
|
assert_eq!(rc, b"CGAT");
|
|
}
|
|
|
|
#[test]
|
|
fn test_canonical_kmer() {
|
|
let kmer1 = b"ATCG";
|
|
let kmer2 = b"CGAT"; // reverse complement
|
|
|
|
let canon1 = canonical_kmer(kmer1);
|
|
let canon2 = canonical_kmer(kmer2);
|
|
|
|
assert_eq!(canon1, canon2);
|
|
}
|
|
|
|
#[test]
|
|
fn test_kmer_encoder_creation() {
|
|
let encoder = KmerEncoder::new(3).unwrap();
|
|
assert_eq!(encoder.k, 3);
|
|
assert_eq!(encoder.dimensions(), 64);
|
|
}
|
|
|
|
#[test]
|
|
fn test_kmer_encoder_large_k() {
|
|
let encoder = KmerEncoder::new(21).unwrap();
|
|
assert_eq!(encoder.k, 21);
|
|
assert_eq!(encoder.dimensions(), 1024); // Capped by feature hashing
|
|
}
|
|
|
|
#[test]
|
|
fn test_encode_sequence() {
|
|
let encoder = KmerEncoder::new(3).unwrap();
|
|
let seq = b"ATCGATCG";
|
|
let vector = encoder.encode_sequence(seq).unwrap();
|
|
|
|
assert_eq!(vector.len(), encoder.dimensions());
|
|
|
|
// Check L2 normalization
|
|
let norm: f32 = vector.iter().map(|x| x * x).sum::<f32>().sqrt();
|
|
assert!((norm - 1.0).abs() < 1e-5);
|
|
}
|
|
|
|
#[test]
|
|
fn test_minhash_sketch() {
|
|
let mut sketch = MinHashSketch::new(100);
|
|
let seq = b"ATCGATCGATCGATCGATCG";
|
|
|
|
sketch.sketch(seq, 5).unwrap();
|
|
assert!(sketch.hashes().len() <= 100);
|
|
}
|
|
|
|
#[test]
|
|
fn test_jaccard_distance() {
|
|
let mut sketch1 = MinHashSketch::new(100);
|
|
let mut sketch2 = MinHashSketch::new(100);
|
|
|
|
let seq1 = b"ATCGATCGATCGATCGATCG";
|
|
let seq2 = b"ATCGATCGATCGATCGATCG"; // Identical
|
|
|
|
sketch1.sketch(seq1, 5).unwrap();
|
|
sketch2.sketch(seq2, 5).unwrap();
|
|
|
|
let distance = sketch1.jaccard_distance(&sketch2);
|
|
assert!(distance < 0.01); // Should be very similar
|
|
}
|
|
}
|