infer_sex

A high-performance, zero-dependency Rust library for classifying samples based on sex chromosome characteristics from summarized variant data.

About

This library uses a multi-pronged algorithm based on chromosome biology. It operates on an iterator of VariantInfo structs in a single pass, ensuring minimal memory usage and high throughput.

The core of the library is the SexInferenceAccumulator, a state machine that you feed variant data into to receive a classification and a detailed evidence report.

Highlights

• Performant: Zero-dependency and processes variants in a single stream.
• Memory Efficient: State machine design uses constant memory.
• Transparent: The final call is accompanied by a detailed EvidenceReport showing the result of each internal check.
• Simple API: A straightforward, builder-like pattern for processing data.
• Robust: Supports both GRCh37/hg19 and GRCh38/hg38 genome builds.

Usage

The primary workflow involves creating a SexInferenceAccumulator, processing VariantInfo structs from your data source (e.g., a VCF file), and finalizing the analysis to get a result.

1. Create a configuration for your genome build.
2. Initialize the accumulator state machine.

   let mut accumulator = SexInferenceAccumulator::new(config);
   

3. In your application, create VariantInfo structs from your data.
4. Process all variants in a single pass.
5. Finalize the analysis to get the result. This consumes the accumulator.
6. Use the structured result.

Algorithm

The final classification is determined by a voting system based on four biological checks.

1. X Chromosome Heterozygosity: Compares the ratio of heterozygous to total variants on chromosome X against a threshold. A high ratio suggests one outcome, a low ratio suggests the other.
2. Y Chromosome Presence: Evaluates the ratio of variants in the non-pseudoautosomal region (non-PAR) to the pseudoautosomal region (PAR) of chromosome Y. A high proportion of non-PAR variants provides strong evidence for one classification.
3. SRY Presence: Checks for any variants within the SRY. The presence of such variants is a strong indicator for a specific outcome.
4. PAR vs. Non-PAR X Heterozygosity: Compares the heterozygosity rate within the X-PAR to the rate in the X-non-PAR. A significantly higher rate in the PAR is a key indicator. Zero heterozygosity in the non-PAR region is also a special case providing strong evidence.

This algorithm has been validated on real-world microarray data.

API Overview

• InferenceConfig: Specifies the genome build (Build37 or Build38).
• VariantInfo: The input struct representing a single variant's chromosome, position, and heterozygosity.
• SexInferenceAccumulator: The main state machine that consumes VariantInfo structs.
• InferenceResult: The final output, containing the final_call (InferredSex) and the report. The call may be Indeterminate when no sex-chromosome evidence is observed.
• EvidenceReport: A breakdown of the results and vote from each of the internal checks.