Example Applications¶

Located in applications/

Sequre includes several end-to-end privacy-preserving applications that demonstrate the framework's capabilities across genomics, healthcare, and machine learning domains.

Overview¶

Application	File	ML Model	Data Type	Description
Credit Score	`credit_score.codon`	Neural network	MPU (partition)	Privacy-preserving credit scoring classifier
DTI	`dti.codon`	Neural network	MPU (partition)	Drug-target interaction prediction
GANON	`ganon.codon`	Classification	MPU	Secure metagenomic classification
Genotype Imputation	`genotype_imputation.codon`	Linear regression	MPP / MPU	Impute missing genotypes across distributed cohorts
GWAS	`gwas.codon`	PCA + linear regression	MPU (partition)	Genome-wide association study with secure PCA
KING	`king.codon`	Kinship coefficients	Sharetensor	Secure kinship estimation (KING-robust method)
MI	`mi.codon`	Multiple imputation	MPU	Secure multiple imputation with Rubin's rules
MNIST	`mnist.codon`	Multinomial logistic regression	MPU	Handwritten digit classification
OPAL	`opal.codon`	Linear SVM	MPU	Metagenomic profiling (secure OPAL pipeline)

Patterns demonstrated¶

`@sequre` entry points¶

All applications use @sequre-annotated functions as their secure computation entry points:

@sequre
def gwas_protocol(mpc):
    X_mpu = MPU(mpc, local_genotypes, "partition")
    ...

MPU with horizontal partitioning¶

Most applications distribute data across parties using MPU(mpc, data, "partition"), where each party holds its own rows:

GWAS: Each hospital/biobank holds patient genotypes
DTI: Each institution holds drug-target pairs
Credit Score: Each party holds customer records

Protocol switching¶

Applications like GWAS and Genotype Imputation use via_mpc to switch between MHE and MPC for operations like eigenvalue decomposition and matrix inverse.

Secure ML pipeline¶

Applications compose Sequre's ML modules:

LinReg → genotype imputation, GWAS
LogReg (multinomial) → MNIST
lsvm_train → OPAL
PCA (random_pca_*) → GWAS
MI (Imputer, MI, MICE) → MI application

Configuration¶

Applications use TOML configuration files in applications/config/:

Config	Application
`credit_score.toml`	Credit score neural network parameters
`gwas.toml`	GWAS dataset paths and PCA settings
`king.toml`	KING kinship parameters
`mi.toml`	MI imputation settings
`pca.toml`	PCA standalone configuration

See Configuration for the full configuration reference.

Loading private data (`collective_load`)¶

The examples in examples/ use synthetic data shared from a trusted dealer, which is convenient for testing but does not reflect how production deployments handle private data. In a real deployment, each party holds its own data on disk and loads it via collective_load.

examples/collective_load.codon demonstrates both protocols in a single file:

Part	Protocol	Type	Application
Credit scoring	MHE	`MPU.collective_load` (partition)	Neural-network classifier on horizontally partitioned customer records
Linear regression	MPC	`Sharetensor.collective_load`	Regression model on additive secret shares of patient data

Run it with:

sequre examples/collective_load.codon --local

For the full guide, see Loading Private Data.