Crate rype 

Rype: High-performance genomic sequence classification using minimizer-based k-mer sketching.

This library provides efficient classification of DNA sequences against reference indices using RY-space (purine/pyrimidine) encoding and minimizer sketching.

Core Concepts

• RY Encoding: Reduces the 4-base DNA alphabet to 2 bits (purines → 1, pyrimidines → 0)
• Minimizers: Representative k-mers selected from sliding windows for efficient sketching
• Inverted Index: Enables O(Q log U) lookups instead of O(B × Q × log M)

Main Types

• InvertedIndex: Minimizer → bucket mappings for fast classification
• ShardedInvertedIndex: Memory-efficient sharded inverted index
• MinimizerWorkspace: Reusable workspace for minimizer extraction

Classification Functions

• classify_batch_sharded_merge_join: Classify with sharded inverted index using merge-join (default)
• classify_batch_sharded_parallel_rg: Classify with parallel row group processing

Modules

c_api

C API for rype - FFI bindings for external language integration.

config

memory

Memory utilities for adaptive batch sizing.

parquet_index

Parquet-based format implementation for inverted index.

Structs

BucketData

Bucket data with minimizers for building inverted index.

BucketFileStats

Per-bucket statistics about total sequence lengths of source files.

BucketMetadata

Bucket metadata for a single bucket.

FirstErrorCapture

Thread-safe error capture that stores only the first error.

HitResult

Query ID, Bucket ID, Score

IndexMetadata

Lightweight metadata-only view of an Index (without minimizer data)

InvertedIndex

CSR-format inverted index for fast minimizer → bucket lookups.

InvertedManifest

Manifest section for inverted index shards.

InvertedShardInfo

Per-shard metadata for inverted index.

LogRatioResult

Result of log-ratio computation for a single query.

MergeOptions

Options for index merging.

MergeStats

Statistics from a merge operation.

MinimizerWorkspace

Workspace for minimizer extraction algorithms.

ParquetManifest

Manifest containing index metadata.

ParquetReadOptions

Configuration options for reading Parquet inverted index files.

ParquetWriteOptions

Configuration options for Parquet file writing.

PartitionResult

Result of partitioning reads by numerator score into fast-path and needs-denominator groups.

QueryInvertedIndex

Query inverted index for merge-join classification. Stores sorted COO (coordinate) entries: (minimizer, packed_read_id).

ShardInfo

Information about a single shard in a sharded inverted index.

ShardManifest

Manifest describing a sharded inverted index.

ShardedInvertedIndex

Handle for a sharded inverted index.

StrandMinimizers

Structure-of-arrays for minimizers on a single strand.

Enums

FastPath

Indicates whether a log-ratio result was determined via a fast path (skipping the denominator classification) or computed exactly.

Orientation

Orientation of a sequence relative to the bucket baseline.

ParquetCompression

Compression codec for Parquet files.

RypeError

Unified error type for the rype library.

Strand

Strand indicator for minimizer origin.

Constants

BUCKET_SOURCE_DELIM

Delimiter between filename and sequence name in bucket sources. Format: path/to/file.fa::sequence_name

ORIENTATION_FIRST_N

Default number of minimizers to check for orientation. Checking the first N sorted minimizers provides a sample for orientation decisions without iterating the full bucket.

Statics

ENABLE_TIMING

Controls whether timing diagnostics are printed to stderr.

Functions

base_to_bit

Convert a nucleotide base to its RY-space bit representation.

choose_orientation

Choose orientation with higher overlap. Ties favor Forward.

choose_orientation_sampled

Choose orientation by checking first N minimizers of each strand.

classify_batch_sharded_merge_join

Classify a batch of records against a sharded inverted index using merge-join.

classify_batch_sharded_parallel_rg

Classify using parallel row group processing.

classify_from_extracted_minimizers

Classify from pre-extracted minimizers against a sharded inverted index using merge-join.

classify_from_extracted_minimizers_parallel_rg

Classify from pre-extracted minimizers using parallel row group processing.

classify_from_query_index

Classify using a pre-built QueryInvertedIndex against a sharded inverted index.

classify_from_query_index_parallel_rg

Classify using a pre-built QueryInvertedIndex with parallel row group processing.

classify_log_ratio_batch

Classify a batch of reads using log-ratio (numerator vs denominator).

classify_with_sharded_negative

Classify with memory-efficient negative filtering using sharded index.

compute_log_ratio

Compute log10(numerator / denominator) with special handling for edge cases.

compute_source_hash

Compute a hash from bucket minimizer counts for validation.

count_hits

Count how many minimizers from a query match a bucket (binary search).

create_parquet_inverted_index

Create a Parquet inverted index directly from bucket data.

extract_batch_minimizers

Extract minimizers from a batch of query records in parallel.

extract_dual_strand_into

Extract minimizers from both strands of a sequence.

extract_into

Extract minimizers from a sequence (single strand).

extract_minimizer_set

Extract sorted, deduplicated minimizer sets per strand.

extract_strand_minimizers

Extract ordered minimizers with positions per strand (SoA layout).

filter_best_hits

Filter a list of HitResults to keep only the best hit per query.

get_paired_minimizers_into

Extract minimizers from paired-end reads.

is_parquet_index

Check if a path is a Parquet-based index directory.

kway_merge_dedup

Merge multiple sorted, deduplicated vectors into one sorted, deduplicated vector.

load_all_minimizers

Load all unique minimizers from a sharded index into a HashSet.

log_timing

Print timing info to stderr if ENABLE_TIMING is enabled.

merge_indices_streaming

Memory-bounded merge with streaming per-shard subtraction.

merge_sorted_into

Merge sorted source into sorted target in-place, deduplicating.

partition_by_numerator_score

Partition reads by numerator classification results into fast-path and needs-denominator groups.

validate_compatible_indices

Validate that two indices are compatible for log-ratio computation.

validate_log_ratio_indices

Validate that two sharded indices are compatible for log-ratio classification.

validate_single_bucket_index

Validate that the index has exactly one bucket and return its ID and name.

Type Aliases

QueryRecord

ID (i64), Sequence Reference, Optional Pair Sequence Reference

RypeResult

Convenience type alias for Results using RypeError.